Leonardo AI for Content Creation: Level Up Your Digital Art & Graphic Design

1. Welcome to AI Art Content Creation with Leonardo.Ai!: Hello, and welcome to the course on Leonardo AI, a Generative content production platform that uses various forms of generative AI, from designing images based on text instructions to video production and beyond. Whether you are an entrepreneur or a startup, looking to craft your first logo or a big company aiming to generate visuals at scale with a consistent style and film, the Leonardo I platform is here to enhance and accelerate your creative process. And if you are a professional designer or content creator, looking to improve your workflow while staying up to date with the latest advancements in AIR content creation, then R the AI is the perfect platform to start with. My name is Anna and I'll be your instructor and mentor for the course. I was the director of product at Series B AI startup based in Singapore, where I relocated six years ago to complete an MBA program. By joining this course, you will get access to over 3 hours of HD video content, step by step tutorials and activities, case studies highlighting real world, practical applications of Leonardo II's generative AI tools, and much, much more this course requires absolutely no prior experience in AI, AI R generation or design. If you are just starting your creative journey with AI, I'll guide you every step of the way. For those who are more advanced in the subject, feel free to look at the curriculum and start with the topics that interests you the most. So let's begin the course by covering how to get access to Leonardo AI and whether the images you create with Leonardo are safe for commercial use. I'll seea in the next video. 2. Account creation and setup. Token system: Everyone. Welcome back. In this video, I'll share details on how you can access the Leonardo AI platform. Leonardo is available as a web app at app.leonardo.ai. Click Create An account and sign up using your email. Leonardo AI is also available for Apple Mobile devices via the IOS app. With an Android app coming soon. In terms of the pricing plans, you have the option to sign up for an individual plan suited for solo creative work, as well as Leonardo for teams. In this video, let's talk about the individual plans. You can choose from the basic free plan and several premium plans. The free plan provides you with a daily limit of 150 tokens and access to some basic features. If you haven't worked with similar IRT generation tool before, you may not be familiar with the token system. Let's talk about how it works. When using the Leonardo platform for image generation, different features have associated token costs based on their complexity and the computing power required. Here is a table provided in the Leonardo user guide that breaks down the token costs per feature, allowing you to plan and manage your usage effectively. For instance, generating an image with default settings of this size will cost you one token. We'll cover all these features later in the course, so it will be much easier for you to make sense of this table. Now, just remember that each time you click Generate to create an image, it costs you a certain number of tokens. If you are on the free tier, your 150 token allowance will reside every 24 hours. Apart from the free tier plan, you can choose from several premium plans that give you access to a much higher number of tokens, as well as the platform's premium features. Please note that for the paid plans, you have access to your full token allowance at any time, and your allowance will renew at the end of the billion period. However, unused tokens expire at the end of the billion period and do not carry over. As I mentioned earlier, apart from the individual plans, we've just covered, Leonardo also provides access for teams. I'll see you in the next video. 3. Is Leonardo.Ai's content suitable for commercial use?: Everyone, and welcome back. One of the first questions you might have when trying a new AI art content generation tool like Leonardo AI is whether the generated images are suitable for commercial use. So let's clarify this right at the beginning of the course. The short answer is yes. However, there are some nuances depending on whether you are using a free or a premium plan. If you generate images under the free tier plan, you do not have ownership rights over the intellectual property of the images, you generate on Leonardo AI. The terms specifically state that the ownership of all intellectual property rights in any output created by free subscribers vests in Leonardo AI, meaning the company retains the rights to those outputs, and because of this, free subscribers would not have the rights to use the generated images commercially. Unless Leonardo AI explicitly grants such rights. On the other hand, if you work under Leonardo AI's paid subscription, you own the intellectual property rights to the content you generate and are free to use it, including for commercial purposes, as long as you comply with the platform terms. As a paid subscriber, you can also select whether your content is public or private. You acknowledge and agree that content marked public will be available for use by all users on the platform. If you opt for content to be private, only you and your authorized users can access that content. As per the terms of service for images kept private on the platform, Leonardo I retains the right to use those images solely for providing the services to the users. You might choose to try out the three tier plan first and then consider upgrading, or you might want to start with the premium plan right away. The decision will depend on your goals for using the platform. I leave a link to Leonard AI's terms of service in the resources section of the video so that you can read it through and decide what works best for you. Okay, great. We've settled all the organizational questions about accessing and using the platform. Now it's time for the most interesting and fun part, creating your first images for this Ilsa in the next section of the course. 4. Introducing Leonardo.Ai's Image Creation function: Hi, everyone, and welcome to the new part of our course. They say, a picture is like a silent story. In this section, you will learn how to create such stories with Leonardo AIs, text to image generation function. We'll start by exploring and getting inspiration from the amazing works in Leonardo Is community. Then you will create your first image from a text description or prompt. I'll walk you through the steps and explain how Leonard Is interface works. We will also look at how you can quickly improve the image generation process. This includes choosing AI models, applying styles and effects to your prompts, and using other helpful tools on the platform. We will also talk about how to craft effective prompts for Leonard AI and what elements to include. You will learn about the platforms features for coming up with and refining your prompts. Towards the end of the section, we will discuss using reference images to help generate stunning visuals, along with your text descriptions. Lastly, you will learn and practice generating images with transparent backgrounds. There is a lot to cover, so let's get started. Sea in the next video. 5. Let’s create your first art work with Leonardo.Ai!: Body and welcome to the first lecture of this section. Let's create our first work with Leonardo AI. When you are just starting out with the new AI art generation tool, it can be quite challenging to come up with ideas for your first image and figure out how to write instructions prompts for the AI to generate the image. That's why the first step I usually take to get familiar with the new software and generate some creative ideas is to open the community page and browse through the works that other users have created. I take note of the styles I like and my want to reuse, and that's what we are going to do in this video. But before we start, let me give you a brief overview of Leonardo II's homepage. After you first log into Leonardo AI, you will see a homepage that is your gateway to various functionalities. Let's explore the navigation bar on the left. Here, you can access your account settings and get information on how many tokens remain available to you. Next, you see a personal feed section where you can browse through all of your previously generated images, as well as other creators images that you liked. Let's return to the homepage. And continue exploring the navigation tab. The next section includes a collection of AI tools that are currently available in Leonardo. You see the same list here under the bring your ideas to life section. These are the tools we will explore in much more detail in the next sections of the course. The next section includes advanced tools that you may want to experiment with, such as training your own EI models. We will cover these tools later in the course as well. Finally, you have access to product updates, the FAQ page, information on how to access premium plans, APIs, and settings. In the center of the homepage, you see a community feed that we are going to work with in this tutorial. This area showcases images created by other users, providing a source of inspiration and a starting point for your own designs. You can either browse through the entire gallery or choose a group that you're interested in. You can also filter images based on their popularity or choose newly created images. Let me choose character here and go through the feed. As the first step, look for images and styles that you like. For example, this one. If you like an image, click on the hard icon on the top right side. This image will be included in your personal feed under the feed tab. Click on the image to go to the detailed view where you can see and copy the prompt that was used to create it, as well as some other details about the settings used when generating the image. We will learn more about the settings you can use when creating an image in the upcoming lecture. For now, please note which model was used to generate the images and what elements were used. You will need these details to follow along with the course. I also find it useful to click on the creators profile and see if they have other works that you would like to add to your feet of liked images. You can follow the creators to keep track of their work. You will be surprised how much you will learn about the capabilities of the platform just by browsing through the community feed without spending any of your tokens. How convenient is that? Whenever you are ready to create your first image, click on the remix button under the detailed view of the image. Click and Remix, we'll copy the prompt and all the generation settings to a new project. Here you can adjust the prompt and change some other settings. Just a heads up if you are on a free plan, you might not be able to remix images that were created using premium paid features. For now, let's make some modifications to the prompt on limb. You can also choose how many images will be created in one iteration. Four are selected by default, and I usually go ahead with these settings. Let's click on Generate. Here, you see a tokens count. To generate an image using this model and the other settings applied would cost me 13 tokens. In the next video, I'll give you a technique on how you can generate great quality images with minimal token usage. So please stay tuned. For now, let's hit Generate. The generation process might take a few moments after which your new images will be displayed. Click on the image to enlarge it and browse through other image variations. Click on Download to save the image to your local drive. For instance, when you want to continue editing this image using other software. All the generated images will also be stored in your personal feed. Let me return to the homepage to demonstrate this. So these are the images we've just created. As always, you can click on the image to get the details about its prompt and the generation settings. If you work under Leonardo AI's free plan, you grant Leonardo AI the right to use, copy, reproduce, process, modify, and publicly display your images. You can read through their terms of use documents for more details. If you don't want anyone else to use your creations, you need to upgrade to a paid plan. Once upgraded, a togle will appear on the left hand sidebar of the image generation function. Make sure that the togle is on if you want to keep your images private. All right, that's it for this video. Now it's your turn to explore the works created by the Leonardo AI community and decide on what image you want to remix. Please share your favorite image in the Q&A section under this video, and I'll met you in the next video. 6. Creating an image from scratch: One, and welcome back. Let's continue exploring the Leonardo AI platform. Now that you have a general idea of what is possible to create with Leonardo, have probably already noticed your favorite AI models and styles. Let's practice creating an image from scratch. Start by navigating to the image generation section from the homepage. You will see a new AI image generation screen where you need to type in your prompt to tell Leonardo what you would like to create. The first thing to decide when thinking about your prompt is your subject matter or who you want to create whether it's a person, an animal, landscape, a fictional character, and so on. Please note that you don't need to add extra instructional words like create an image or imagine at the beginning of your prompt. Let's start with the short prompt like futuristic SEM. The next thing you must choose before starting the generation process is the AI model you will use for generating an image. The model selection is available under presets on top of the left side bar. I recommend you start working with the lighting models, which are one of new offering from Leonardo designed to speed up the image generation process while maintaining high quality outputs. In addition, the models are very cost effective. For instance, generating an image with the Leonardo Phoenix model would cost 24 tokens per four images. While if I switch the model to Leonardo lighting, the generation cost drops to just ten tokens. Now, you have two specialized versions of the Lighting models, which include Leonardo Lighting for the realistic images, and Leonardo Anime for anime cartoon style images. For my work, I'll choose Leonardo Lighting. Let's also explore the settings on the left hand side bar. Pick between fast and quality generation mode. First comes with reduced generation time at the expense of quality and resolution limits. Quality mode comes with slower generation times, but image detail and quality will be significantly improved. The quality mode is optimal for big aspect ratios and high resolution images. Please note that if you choose quality, the talking count for the generation will be increased. Next, you can choose image aspect ratio and also set image dimensions. I'll choose landscape and large. Here, you can modify the number of generated images. I'll stick to four the default settings. Now click on Generate to start the creation process. Leonardo AI will interpret your prompt and start rendering the images, which may take a few moments depending on the complexity. Next, you can add extra details to your prompt. We'll have separate lectures focused on how to create effective prompts for Leonardo. But for now, let me say that I recommend starting with a simple prompt and then adding extra details and descriptors one by one to see how the changes influence the output. Let's come back to the demo. Mom. Ah. I like this image best. So let me click on the download icon to save the image to my local drive. In case if you don't like an image, you can always delete it by clicking on the win icon. Remember that all the generated images will be available for you in your personal feed section. Let's return to the homepage and go to the library. And here you go. Alright, that's it for the video, and I'll see you in the next one. 7. What is an AI model, and how to choose the right one for your work?: Everybody, and welcome to the new lecture on using Leonardo AI for AI art Generation. In the previous video, we briefly talked about selecting an AI model that will be used to generate an image. This model will have a major impact on the quality of the final output. So the decision on which model to choose is important. Let's talk about what an AI model is, what models are currently available in Leonardo. How to choose the right one for your work. Let's start with the definition first. AI model in the context of art generation is a trained system that can generate images based on textual or visual inputs. These models are trained on large datasets of images and correspondent descriptions to learn various styles, objects, and scenes. The role of an AI model in our generation is to interpret the input prompts and produce creative coherent and visually appealing outputs that meet the user's requests. By the way, if you would like to learn more about AI fundamentals and how AI creates art, there will be a separate section in the course to cover this, so don't miss it. Now let's talk about the types of models available in Leonardo. The platform offers several types of models, general based models, such as the SDXlO Stable Diffusion 1.5 and 2.1 models. The next category includes fine tuned models which are specialized versions of the general models, but adjusted to perform better on specific types of images or styles. For example, if the fine tuned model is strained on landscape photography, it will produce more detailed and accurate landscape images than the general model. There is a new family of fine tuned models called Leonardo I Lighting Excel models, which we briefly covered in the previous video that I would like to mention here as well. These models are new offerings from Leonardo optimized for speed, allowing for faster generation times, which is crucial when you need to produce large volumes of images quickly. However, the speed optimization doesn't compromise the quality of the images. You will still get high quality outputs with these new models. If needed, you can further upscale them with universal upscaler Leonardo AI module that we will cover in the following sections of the course. Finally, you can also train your own models to get outputs tailored to your unique style and preferences. This is considered a more advanced technique and not something you will start with immediately, but we will be covering it as well. If you now feel a bit lost with such a variety of models and think it's very challenging to choose the one for your project, here is the process that I recommend you follow. Start by using Leonardo Lighting models such as Leonardo Lighten Excel for photo realistic images, and Leonardo Anime Excel for cartoon style images. Go through the images that you saved two favorites from other creators and notice what models they used in their generation. For example, this is how I discovered my favorite model so far, Leonardo Vision Excel. I noticed that this model has been used by nearly all the creators whom I follow and whose work I like. So I started applying it to my projects as well. Another technique that I like to use is going through every model one by one and testing it with the same prompt. This will give you a good idea of what each model is capable of and what your preferred models are. Yes, this technique will require you to spend quite a bit of tokens. But even if you are now working on a free plan, you can still do these tests. Let's say by checking one new model every day and using the remaining tokens to follow along with the course. Remember, your talking scount will reset daily in case if you work on a free plan. Please let me know in the Q&A section of this video, which model is your favorite. And that's it for the lecture. To sum up, AI models are trained systems that generate art from textual or visual prompts. Leonardo supports several types of AI models you can choose from, such as base models and fine tuned models that are specialized for better performance in specific styles or themes. Light and Excel models offer fast generation times and cost efficiency suitable for high volume image creation. These are the models you can start with first and then expand to other models you've tested and like the best. In our next lecture, we will look at other parameters that influence image generation. I'll see you there. 8. How to improve your images quickly using Presets and Elements: One and welcome back. In this video, we will explore how to influence the image generation process with Leonardo II's presets and elements features. These tools allow you to apply complex artistic styles and adjustments with ease greatly simplifying the image creation process. Let's get started by talking about presets first. Presets in Leonardo AI are predefined settings that can be applied to image generation requests to quickly achieve specific styles or effects. These presets can adjust aspects like color schemes, mood lighting or whether your image resembles an illustration or a photo, providing a quick way to apply complex artistic transformations to your image. Let's see how we can quickly transform the further realistic image of the futuristic city from the previous example into an illustration. Preset selection is available from the drop down list on the left hand side bar of the image generation page. I'll choose sketch, color, and then click Generate. We see obvious changes in the style of these new images, and they clearly resemble sketch right now. Let's try several other presats. All results look great, but for this work, I prefer colored sketch. After experimenting with and trying different presides from the list, you will find the best match for you and we'll be able to quickly choose them for every next image you create. Apart from presets, there is another feature that can help you give additional instructions to the AI model on what style and visual outcome you are looking to achieve. These are called elements. You can add elements from the AI image generation page by clicking on the icon from the left side of the prompt bar. And from here, go to elements. Browse through the collection of elements. The element thumbnail and description will give you a general idea of what style modifications you should expect after applying this element to your project. Some elements look best when applied with a specific model like Leonardo vision. You can find this information in the elements description. It is recommended to add up to two elements into your prompt, as adding more elements may lead to unexpected results. All right, let's try different prompt for a change. I'll choose Leonardo vision model. The preset style is selected as dynamic by default. So let's keep it. And first, let's generate an image without applying any elements. I'll hit generate here are the generated images. They look great but could definitely be improved. Let's open Elements menu. And from here, select psychedelic art. Let's hit Confirm. Notice that the element's thumbnail appears at the bottom, just under the prompt bar. The impact of each element can be adjusted using the strength slider, providing you find control over the influence each style has on the final image. For the first iteration, I leave the weight unchanged and then click Generate to see how the elements will blend into the creation. Wow, the images look incredible and notice the drastic difference between these images and those we generated the first time. Now, let's reduce the strength and click Generate as well. For this work, I think that the higher weight is better, but I recommend you experiment with the setting for every element you apply. Let's add other elements to see how they would modify the image. Mom, Dom. 9. Organising your work into Collections: One and welcome back. Now that we've covered the basics of starting the image generation process with Leonardo, let's talk about how you can organize your generated content. For this, you can rely on collections. In other words, folders. There are different ways you may structure the collections. You can create them based on your current projects, various teams in your organizations or by the type of assets you are creating. Similar to how we organize folders with files on a computer, you can create a hierarchy of levels within a collection. For instance, I organize my collections based on the sections of this course. I have one collection per section. Inside each section, I have sub collections relevant to the lessons of the section. Let's see how to create a collection and add images to it. From the Leonardo homepage, go to library and from here, click on collections. Then choose new collection. Type in the name for your collection and then click Create and add Images. After you've selected images that you'd like to add to your collection, click on Confirm. Now, when you look at your personal feed, you will see that images are organized nicely into the collections. Let's create a collection with several levels. I'll drill down to the course Section one collection. And from here, click on New collection again. I'll choose Create and add Images. Let's choose these two and click Confirm. So we've just created a hierarchy within the collection. You see that if I drill down in the course Section one collection, there will be a new one called Lecture One. All right. Let me show you another option for how you can add your content to the collection. Let's return to the personal feed. From here, click Select Images. Choose the images you'd like to add to the collection. And click Organize. From here, choose the folder where you want to move the images. I'll choose Lecture one collection under the course Section one. We see that two images have been added to the Lecture one collection. Great. In case you'd like to delete a collection, return to the collection tab. Choose the collection you want to delete and click on the three dots icon. From here, select Delete and make sure that in case if you'd like to delete the collection only without the source images, the delete images everywhere checkbox remains unchecked. And next, click on Delete. All right. And that's it for this tutorial on how to organize your content, and I'll see you in the next video. 10. Creating Photo-Realistic Images: Everyone, and welcome to the new video of the course. So far, we've covered the following tools available for you to influence the image generation and tailor it to your vision. Selection of AI models, including Leonardo AI's special fine tuned models, variety of presets and elements that you can add in addition to your prompt description. However, these are not all the tools available for you. In the upcoming lectures, we'll continue exploring other settings you can rely on when generating images with Leonard AI. In this lecture, we will talk about the for the real function, which as the name suggests, can significantly enhance the realism of generated images. With this powerful tool, you don't need to design overly complex and nuanced prompts as the Foer real function will do the hard work for you. Let's see how it works. From the Leonardo AI homepage, go to the image generation page and then go to the advanced settings. Here, activate the Forter real function. For this tutorial, I'd like to choose a different model for a change. Please notice that the model selection is available directly in the advanced section settings. Let's choose Leonardo Kino Excel as I like best how it performs with the for the real function. Let's type in the prompt. Let's hit generate. I really like the results. The images look like for us. They are very realistic and create the impression that we are present in this art studio watching the artist at work. Leonard Duy has done an amazing job. As always, we can choose different presets and also select elements to add extra flavor to our work. Let's try cinematic close up preset. Let me also choose another preset fashion and see how the images will change. It's interesting how preset style can drastically change our image. You see that when I change the preset to fashion, we get these beautiful women characters in all the pictures and the atmosphere of the images has changed as well, becoming more serene, graceful, and delicate. Et's experiment with more presets and elements. Mm hmm. I think these images look incredible and notice the drastic difference between the images that we've generated without using the elements. Please feel free to experiment with adding presats and elements together with the photo real function for your own project. And that's it for the tutorial and Alca in the next one. 11. How to enhance your generated images with Upscaling and Refining: Hello, and welcome back. So far, we've talked about how you can influence the image generation process with tools such as presets, elements for the real however, oftentimes you may want to make some adjustments to already generated images, and Leonardi provides many possibilities to do this. In this lecture, let's talk about some of the easiest methods to upscale and refine your image. Before we start the demo, let me clarify the terms. Upscaling refers to the process of increasing the resolution of an image. When you upscale an image, you are essentially enlarging it from a lower resolution to a higher resolution. The goal here is to make the image larger without losing clarity or introducing blurriness. Refining an image involves enhancing the quality of the image in various ways. It can include improving sharpness, adjusting color balance, enhancing detail, and reducing noise. Start by selecting an image from the generation history feed in the image generation function. Next, click on the upscale button. Here, you can choose between Ultra and Alchemy modes. Let's cover them both. Ultrapscaler is a new feature from Leonardo team that works extremely well in adding fine details and is the recommended upscale method. Just the parameters. Upscale style. Choosing between artistic and realistic will greatly affect the result. Leonard recommends selecting artistic when using ultra with two D style imagery. Let's choose artistic. Upscale multiplier determines how large the image will be upscale. Now let's open advanced settings. Creativity strength. It controls the level of creative variation introduced when adding extra details to the image. Keep in mind that higher strength settings can significantly alter the image from its original form. Let's increase it for a bit. Details contrast. This setting adjusts the contrast of details within the image, much like high dynamic range options in other upscale services. It can create a strong effect, particularly along the edges of objects and may appear unrealistic if set too high, similar to the extreme unsharp mask effect in Photoshop or the structure sharpness settings in Photo editors. For my work, I'll keep details contrast without changes. Finally, the similarity settings, it determines the extent to which the overall structure of the image stays similar to or deviates from the original. The setting is particularly useful when combined with creativity strength to maintain some resemblance to the original image. Let's increase similarity here for a bit. After you are done with the settings, click on scale to confirm. Notice that the token cost will be displayed in the bottom. Et me download the upscaled image as well as the original image so that we can open both of them to see the difference. We have the upscaled image on the left hand side, and to me, it looks brighter and crisper rather than the original image on the right hand side. Let's close the images and continue our demo. As I mentioned, ultra is the recommended upscale mode, but to make the tutorial complete, let's cover the other available modes. This time, I would like to upscale this image. I'll click on Upscale image from the image generation page and from here, let's go to Alchemm. Alchemy upscale is much more simplified than ultra and acts more as an image refiner than proper upscale. It is ideal for quickly refining images, adding subtle restyling or enhancing their definition. As you can notice, it is not possible to adjust the amount of visual changes that will occur in the result, unlike in ultrascalar. Let's start by selecting a refiners strength. Higher the strength, the more the refiner will adjust the image, attempting to improve it while upscaling. If you are fine with the original image, use the medium or low setting. For the purpose of demo, let me choose high. You can also toggle smooth mode, which enhances image coherence and improve hands and faces. Once you are done with the settings, click on upscale. Once the process is complete, you can compare the original image with the upscaled version by choosing between them on the lower left of the image. Let's also download both images and compare them. The image that was enhanced with Alchemy refiner is on the left hand side. It looks smoother than the original image, likely because I've selected this mode as the refiner setting. Personally, I like the original image best, but please let me know in the comments in the QNA section for this tutorial, which of these two images you like best. Let's continue the demo. Alternatively, you can also access the upscale function from the homepage. From here, go to your personal feed, select an image you'd like to modify and click on upscale image. You will see the same screen for choosing between the ultra and alchemy modes. We've just covered. You can easily access all of your upscaled images by selecting the upscaled tab from your personal feed page. Please note that this is just one method to upscale and refine an image. We will discuss other options available in the next sections of the course and I'll see you in the next video. 12. Top 5 recommendations for creating a great prompt: Everyone. Welcome back. As you saw from the previous lectures in Leonardo AI, you have a variety of tools to influence the image generation process, like selection of models, presets, elements, and much more. But regardless of whether you choose to use them or not, there is one thing that you have to come up with to initiate the generation process of your visual it's your prompt description. In this video, I'd like to share my top five recommendations for creating a good prompt one that together with other tools, helps you to get high quality visuals that are closely aligned with your original vision. Let's get started. So the first recommendation, go through this prompting checklist to decide on what you'd like to create. Do you want a photo or illustration? The answer will help you to choose between the graphic focused or photorealistic models that will be used for image generation. What is your subject matter? Person, animal, landscape, fictional character, and so on. What specific details and effects do you want to include? Please don't feel stressed by all the details and nuances you need to think about to make a good prompt. First of all, you don't need to include everything from this list. Think of it as a framework that can be adapted or simplified depending on your specific needs and the level of detail you want to achieve. I recommend always starting with the simple prompt and then adding extra details one by one to see how they affect the image. Okay, let's come back to the prompt check list. Include extra descriptions in your prompt to change how the AI creates the image or to add more special touches to it. Here are just some examples of what you can add to the prompt type of photography, emotions and moods, magic words, specific art styles. And by the way, in case you'd like to mimic the style of certain popular individuals, companies or studios, you can do so by referencing them in your prompt description. All right, so far, we focused on what elements to include in your prompt. However, please note that apart from the elements themselves, you need to think about the order you place them in a prompt. So the words at the beginning of the prompt have more weight than those at the end. Let's compare these results. Moving next, recommendation number three. For detailed prompts that have multiple elements include commas to separate ideas. This helps the EI model to understand and process each detailed better. Recommendation number four, be practical when creating prompts and avoid designing overly complex and unrealistic prompts, as they may result in poor or ambiguous outcomes. When in doubt, go for shorter prompt descriptions and see where AI takes you. You can always tweak the prompt later on or add additional effects through other tools like elements or reference images. Let's move to recommendation number five. Remember that you can add negative prompts for the cases when you want to exclude undesired elements from your image like text, numbers, certain colors, et cetera. Click on add negative prompt and describe what you'd like to exclude in the negative prompt field. In case you added a negative prompt, but are still unsatisfied with the outcome, consider experimenting with changing the model aspect ratio, presets and elements there are other tools available for modifying already generated images, and we are going to cover them in the following sections of the course. Okay, and that's it for the five recommendations on how to design your Leonardo EI prompts. As always, let's recap on what we've just learned. Review the prompting checklist to decide on what you'd like to create. This includes deciding if you want a folder or an illustration, your subject matter, specific details, and effects you want to include, as well as extra descriptors. Pay attention to the order of the words. Those at the beginning have more weight than those at the end. Include commas to separate ideas in detailed proms with multiple elements. Avoid designing overly complex and unrealistic proms as it will be difficult for AI to interpret them, and finally, add negative proms to exclude undesired elements like texts, numbers, and others. That's it for the lecture. I'll meet you in the next one where we talk about some prompt ideation techniques. 13. Improving and ideating your prompts with Leonardo.Ai: One, welcome to the new lecture of the course. Now that you know how to structure your prompt to achieve the best outcomes, let's talk about several other tools available on the Leonardo AI platform for ideating and improving your prompts. To access the prompt ideation and improvement tools, click on the icon on the right hand side of the prompt bar. The first tool useful for exploring the platform's possibilities in the early stage is the new random prompt feature. This tool is especially handy when you need a fresh idea or want to break out of a creative block to generate random prompt, click on the die icon. I often use this tool to get fresh ideas and new perspectives on what I can create with Leonardo. Definitely check it out. Let me click on Generate to see this prompt in action. The second tool you can try is the improved prompt feature. It helps refine a basic prompt into a more detailed and complex one by adding necessary information or suggesting alterations. This feature works for prompts that are less than 200 characters. For example, you can start with a very short prompt like foreign flowers. And then click on Improve prompt. This feature gives us a highly detailed prompt, something that might be hard to come up with from scratch. As always, you can adjust the prompt and the image generation settings before clicking on Generate to see the outcome. All right. Moving on. Edit with AI is a handy feature that allows you to instruct the AI on specific changes you'd like to make to the prompt. For instance, let's type in the same prompt, foreign flowers and ask AI to expand it and make it a sketch. Leonardo AI provided prompt modifications closely aligned with our instructions. Great work. Now, let's generate the image. Awesome results. And let's continue our demo. The fourth useful feature I'd like to cover is describe with AI. Sometimes you have an image or images that inspire you and you want to use them as a reference. But coming up with a good prompt description from scratch can be difficult with the describe with AI feature, you can submit a reference image and get an initial prompt to start your creative process. Let's see how it works. For the demo, I'm going to submit this image of Tesla Cyber track as a reference. And we've got some decent image description that we definitely can begin with. This describe with EI feature I missed when using other EIR generation tools, so kudos to the Leonardo Team for introducing it. Alright, cool. Please share in the Q&A section which prompt improvement feature is your favorite, and I'll see you in the next tutorial. 14. Improving your images with Reference image: Image-to-Image reference: Everyone, and welcome back to the lecture. In this and the next several videos, let's talk about using reference images in the image generation process. When you upload an image or images as a reference, you instruct AI models to create a new image based on the look and feel of the reference image. Leonardo provides different options for you to specify how the reference image will be used. The option available in the free plan is image to image. This parameter detects the color pattern and the overall look of a reference image and uses this to guide your image generations. Let's see how it works in action. Open image generation. And from here, proceed to the image guidance tab, which becomes available after you click on this icon on the left hand side of the prompt bar. Choose image to image and click Conform. Next, you'll need to upload the reference image. This is the image that I'm going to upload. You can also modify the strength parameter that sets how much of the reference image will be applied to the newly generated image. For my work, I'll leave the strength parameter without changes. Last but not least, make sure that the aspect ratio of the reference image is compatible with the aspect ratio of your generated image for the best results. Everything is good for my case. For this work, I'll use the model Leonardo Lighting. Let's modify the prompt. I'll type cyber punk, C girl, and I'll leave all other parameters without changes. Let's hit generate. Nice results. However, I'd like to have more resemblance between the generated and original images. For this, we need to increase the strength number. Let's do this. I'll increase it to 0.7 and hit generate one more time. Great results, and we see that the new images definitely look more like the original reference image. But here we have another problem. Now, we don't see enough of the cyberpunk style from the prom description, and this is quite logical. By increasing the strength parameter, we ask the AI model to put more emphasis on the reference image rather than on the prompt. There are a couple of things you can do to improve this. First, you can explore elements and apply those that might work for your project. You can also play with changing the model and see if you can get better results. In case the model has been trained on similar images to what you'd like to create, cyberpunk style. In my example, you will get great results. Now, in case you signed up for the premium plan, you have many more options to improve the style of the image while keeping the resemblance with the original image. We will explore them in the upcoming videos. I'll see it there. 15. Improving your images with Reference image: Style reference: Everyone. Welcome back to the video where we continue exploring how to use reference images to impact the image generation process with Leonardo AI. In this video, we will talk about premium features that give you much more control over how the reference image will be used by the AI model. If you are not using the premium plan yet, you can still watch the lectures to be aware of what's possible with the extended functionality of Leonard the AI. Let's get started. In addition to the image to image option, we covered in the previous tutorial. Premium plan subscribers have access to other options for using a reference image, such as style reference, content reference, depth to image, edge to image, pose to image, text to image. Leonardo AI keeps introducing new options for using reference images. So stay tuned and check their user guide for updates. I'll leave a link to the guide in the resources section of the video. It's better to see something once than to hear about it 1,000 times. Let's jump straight to a demo to see examples of using the reference image options. To recap, we'd like to retouch an existing image so that it has a cyberpunk look while recreating the exact likeness of a person in the reference image. We uploaded the first reference image. Next, we set its type to image to image and increase the strength to 0.7. Now let's upload a second reference image. But this time, let's choose a different type style reference. You can choose a reference image from this gallery or upload the one from your local drive. I already uploaded my reference images before, so I'll choose this one and then click Confirm. With the reference image set to style reference, the e model will take the aesthetic qualities or style from the reference image and apply it to newly generated images. Make sure that the aspect ratio of the style reference image and generated images are compatible. You can also modify the strength of the style reference. For this iteration, I leave it unchanged. Another thing to keep in mind is that when using the style reference, Leonardo recommends using regular stable diffusion models instead of lighting models for better quality. My test shows that diffusion Excel model works best for the cyberpunk effect. However, please note that the stable diffusion model cannot be selected from this list of presets. Instead, go to advanced settings. And choose the model from this list. Let's also switch to quality generation mode for the best quality images. And now let's hit Generate. We see some nuanced elements applied to the image here, but let's try to increase the strength to see if the results can be improved. We can also upload a second reference image by clicking Add more images. As always, you can either choose an image from the gallery or select the one from your local drive. Let me select this second reference image and click Confirm. For multiple style reference images, all inputs will use the same overall strength. The amount of influence each individual reference image has can be adjusted with the influence slider. I leave the setting as for this iteration, and let's click Generate. We are getting closer to the cyberpunk style. We can continue our experiments by playing with the strength slider of the reference images. And perhaps we can lower the strength of the image to image reference image. Interesting results. Right now, you can upload up to four style reference images. So for the purpose of experiments, let's upload the third one. This time, I'll choose it th my local drive. In addition, you can also add elements. When using elements together with style reference, it is recommended to increase the strength of the elements more than usually needed. I could not find any elements that would work well for my project, so I'll press Cancel here and let's hit generate. We've got some interesting results, but I would continue playing with the strength of the image to image as well as style reference to get the best outcome. That's it for this video, and I'll meet you in the next one. 16. Improving your images with Reference image: Content and Post-to-Image reference: One. Welcome back to the series of Tutorials, where we experiment with using reference inputs for the image generation process. In the upcoming video, we will also restyle our existing images using other types of references called content reference and pause to image. Content reference transfers the general shapes of the reference image, typically without transferring colors or textures. Post to image reference is quite self explanatory. It will analyze a reference image, look for human or similar figures and try to reproduce their poses for the newly generated images. Let's open Leonardo AI to see the demo. Let's choose content reference. And here select this image that we are going to use for the demo. At click conform. Let's change the prompt field. I'll type three D cartoon style Image of business woman. We're in A cheek, a white blazer or dark. Top bear it with like pins. We will continue using the Leonardi diffusion Excel model. Please note that similar to image to image reference, it is recommended to use regular stable diffusion models instead of lighting models for better quality. Since my reference image has landscape resolution, let's also select landscape dimension in here and let's hit generate. Amazing results. The new images have similar shapes as the reference image. Notice that the lady in the foreign round has a similar pose and outfit as in the reference. And we have a very similar background with the cars in the city center. However, please note that the content reference may not work well for creating the exact likeness of a person or character. So consider using image to image reference instead. Alright, let's try another prompt. If you like, you can choose the strength of the content reference image as well. I'll leave it at high. I'll keep the model Leonard Diffusion Excel, and let's click Generate. Okay, great results. And let's also change a model to Leonard de kina for a change. I'll click Generate one more time. Please also note that when using content reference, you should avoid using elements as they will have extremely limited effects on the final results. And here we go. We see that the cyberpunk style is much more visible as opposed to the example from the previous lecture. In case we have a style reference image we want to use for our image generation, together with content reference, we can edit as well. I'd like to experiment with the style of this image, let me choose it as a reference. I'll type in a new prompt. Let's keep the model Leonardo Kina Excel. And for this project, I'd like to activate photorealFunction. Now let's hit generate. Great results, and I think they look very realistic like the real for us. Now, let's try pose to image reference. Before we start, let's remove these two references and then choose pose to image. Let's select this reference image for a change. Let's also change the aspect ratio. I'd like to change the model to diffusion Excel. And let me change the prompt. I really like these results. We see quite similar polls as in the reference image without any background details. As always, we can play with the strength slider of the reference image. We can also modify our prompt if we would like to bring any modifications to the first iteration. With the increase of the strength of the reference image, the resemblance of the new image to the reference one became more obvious. And now we also have a neon color on the background. Please feel free to experiment with these reference images that we've just covered and I'll see in the next video. 17. Improving your images with Reference image: Content reference for creating texts: Everyone. Welcome back. In this tutorial, I'd like to show you another example of using an image as a content reference. I have this image of a large text that I prepared in Canva and I'd like to use it as a content reference and transform it into a highly stylized output. Let's do this. So I'll choose content reference. Then I select an image from my local drive. I'll type in my prompt. As for the model, let's choose Leonardo Diffusion Excel. It is available in the advanced settings. And here we go. Let's choose an aspect ratio to the landscape and switch to the fast generation mode. I'm fine with other settings. I'll click on Generate. Awesome results. The output is unique and artistic. Splashes and drips extend from the leathers as if they were freshly painted and still wet. As you remember, Leonardo recommends avoiding using lighting models together with content references. But let's experiment and see what results we'll get. So instead of Leonardo Diffusion Excel, let me choose Leonardo Lighten Excel. And I leave all other settings without changes. The results are okay, but not as good as when we used non ten model. Let's do another experiment and change the reference image settings to text image input. I'll click View More and select text image input here and click Confirm. I'll choose the same reference image. Text image reference type allows you to generate stylized text art as well. Let me remove content reference and let's keep the settings without any change. Let's hit generate. Interesting results, and I really like the look of the background color. And since we are experimenting, let's return to the Leonardo Diffusion Excel model and generate one more time. I like the look of these bold, three dimensional letters and the color gradient. The last experiment that I'd like to run for this demo is changing the generation mode quality from fast to quality mode. Notice that the token counts has been increased to 38. Please bear this in mind when changing to quality generation mode. Great results. Personally, I like these outputs better than the previous one. But let me know what's your choice in the Q&A section for this video. And that's it for this tutorial, and as always, Alca in the next one. 18. Improving your images with Reference image: Creating your own image reference: One. Welcome back. So far, all of the reference image examples were about using an existing image as a style reference. In addition to this, you can generate an image with Leonardo and then choose it as a reference for your next work in just one click. This is especially useful when you need a final image with a certain texture or color pattern, apart from experimenting with generating reference image, we'll also try out two other reference types edge to image and depth to image. So we have a lot to cover. Let's open Leonardo to start the demo. Let's begin by generating our reference image. I already prepared my prompt. Which is marble texture of different iridescent colors. I'll keep the model as Leonardo Lighting and image dimension as a landscape. Let's click Generate. I very much like this texture, and I hope it will work great as a style reference image. So let me download one of these images. I think I like this best. And the next thing I need to do is to choose it as a style reference image. So I'll click Style Reference and select this image from my local drive. Now let me change a prompt. Let's keep all other settings without any changes and hit generate. You see here that the style has been applied to both the foreground and the background of the image, not just to the kitchen furniture, and that's not the effect that I was looking for. Here is what you can do to fix that. Let's remove this reference image and generate a new 11 more time. So here is a prompt that I used to generate the first reference image. But this time, let me add these words, three D, material, sphere, on a white background. I added three D material sphere on a white background so that the texture I want to create is applied to the main subject of my original image, like the kitchen furniture in my project and not to the entire image. Let's also change the image aspect ratio to square one. Otherwise, we might get two or even more spheres, and let's hit generate. For our purposes, this image will work best as we have a three D sphere on a wide background. So let me download this image to my local drive. And now let's use the new image as a style reference. So I'll return to my original prompt. M. Still not quite perfect as we see some parts of the texture on the image background. Let's make an experiment and change the aspect ratio to portray and heat generate again. The results are pretty much the same as with the square image output. So the next thing that we can do here is to either reduce the strength of the style reference image or change the model. Let's try these options. These results are pretty good, let me also change the model. I'd like to test Kino as well as vision. These results are also amazing. And notice that we have different pitch and outlook when we are changing the model. And let's try lifelike vision as well. Et's continue and add a second reference image. This time, I'd like to add edge to image reference. Edge to image is effective at replicating the composition of an image. Here is an image that I'd like to use as edge to image reference. So let me submit it to the platform. Let's switch to the lighting model again and click Generate. Great results, and we definitely see the resemblance in the composition of the kitchen furniture when compared to my reference image. As always, I encourage you to also experiment with changing the model to achieve the best results. Okay, cool. Let me also show you another type of reference image. Depth to image type uses depth information to enhance the three dimensional aspects of an image. So I'll remove edge to image reference and go ahead with choosing a new one. Let's use the same image as a reference. And I'm fine with all other settings. Let's hit Generate. We don't see much difference here if we compare this output with the previous one when we used edge to image reference. Apart from experimenting with changing the model, don't forget that you can also change the strength of each of your reference images. Please feel free to share your creations by posting links in the Q&A section for this video, and that's it for this tutorial, I'll seea in the next video. 19. How to Create an Image with a Transparent Background: One. Welcome back to the video on Leonardo AI text to Image function. This time, we will learn about yet another capability, creating transparent images. Often, you need to remove the background of your image. For instance, when you want to use it as part of another composition or when you need to change the background, Leonardo provides a background removal option for all the images that you generate. To remove the background, just click on the remove background icon available in the bottom toolbar of your image. The background removal works best if there is a distinct object located in the foreground of your image. However, even in this case, you may notice some imperfections and parts of the old background here and there. But here is what you can do instead. You can guide Leonardo to generate the image without a background from the get go without the hassle of removing the background after the image has been created on the homepage, click on Image creation under bring your ideas to life. Go to advanced settings and enable transparency. Next, enter your prompt. Please note that shorter prompts work best when the transparency feature is on. Now, in terms of the model selection, according to Leonardo's documentation, the transparency feature works best with Leonardo Kino vision, and Albedo base. Let's choose Leonardo Kino. Let's also change an aspect ratio to square, and I'll hit generate. That's an interesting issue that we've got. It seems that the transparency feature was switched off when I changed the model to Kino. Let's check if my guess is right, so I'll activate the transparency one more time. Then I changed the model to vision. And let's check if the transparency feature is on. No, it's not. So the first thing we need to do is to decide on what model we'd like to use, and after that, activate the transparency feature. Please keep this in mind. Let me return to Kina. Switch the transparency on. All other settings seem to be fine. Let's try one more time. This time, the background is transparent, and it highlights the squirrel's geometric low poly art style design. If you like, you can also add elements to your project, but not all elements are compatible with transparency. I recommend you check this table available in the user guide to see if the model and the element can or cannot be used. I'll leave a link to this guide in the resources section for the video. Let's see which element gives the best result according to the user guide. I'll choose Color Pop. Lowering the strength of Elements usually improves the output. So let's do this. Let's make sure that the transparency feature is on Allgood here and I'll click Generate. The color schema in the second generation is richer and includes a wide range of hues from deep oranges to soft pinks and purples. Compare it to the more monochromatic orange and white scheme of the first squirrel. Let me know which of these two generations you like best in the Q&A section for this video and that's it for the tutorial, I'll meta in the next one. 20. Introducing practical use cases for the Image Creation function: Everyone, and welcome to the new section of the course. So far, we've talked a lot about the various capabilities available in Leonardo AI to create images from text descriptions. In this section, we'll focus more on the practical applications of these tools and see how you can use them for business purposes such as marketing material creation, product prototyping, UX, UI design, logo design, creating interactive content, and more. I include these use cases to give you even more ideas on the practical application of this new technology and inspire you to go beyond this course, incorporating the tools you've learned into your daily work routine. I encourage you to follow along with every use case that you are about to see. Please refer to the PDFs attached to each lecture with examples of prompts you can try as well as additional resources you may need to follow along with me. Of course, the use cases covered in this section are by no means all that Leonardo AI can be used for, I would like to ask you to share in the Q&A section of the video how you are using the product. I promise to review all the answers and include additional tutorials covering the most insightful or unusual cases. And without further ado, let's begin. 21. Use Case 1: Marketing - Creating assets for social media posts: One and welcome back. So the first use case we are going to explore is using Leonardo AI for creating marketing materials. Imagine you work as a social media manager for an eco friendly tour company that offers guided nature experiences. You'd like to create captivating social media content to inspire your subscribers to start planning their next trip, and you want this content to showcase the scenic beauty of your locations without investing in complex and expensive location for the shoots. You can use Leonardo AI to create images for your company's Instagram feed, stories, and real covers from the homepage. Open Image creation function. Next, type in your prompt. I've prepared this one. Let's choose a model lifelike vision. Choose between fast and quality generation modes for Instagram posts, select square aspect ratio. And for my work, I'll stick to four generated images. Let's click Generate to see the first results. We've achieved very decent results with the first iteration. Here is what you can do to improve them even further. First of all, you can experiment with choosing different preset styles instead of standard dynamic one offered by Leonardo AI by default. You can also choose one or several elements from the collection of elements offered by Leonardo. And if you generate photos like in my example, you can enable PhotoealFunction for more realistic images. To do this, go to advanced settings and activate Photo real. Let's see what results we'll get this time. Notice that the number of tokens required for this generation has been significantly increased since when you select FOA Real, Leonardo switches from fast to quality generation mode. So please keep this in mind when working with FOA Real. Awesome results. The colors are very vivid and all the pictures are very photo realistic. Oftentimes, you need to generate images in a specific style unique to your brand. Let's say that for your next set of images, you'd like to mimic this aca friendly miniature photography style. So let's use this image as a style reference for the next project. I'll first change my prompt to a new one. And as we agreed, let's add an image as a style reference. I'll leave all other settings unchanged and at generate. I like this miniature style. It is very cute. You can change the strength of the style reference from medium to high if you would like to increase the style strength, or you can add a second image as a style reference. For the demo, I've prepared this image that I'd like to add to my project as well. Let's do this and see what changes it will brings us. Here is an image And actually, let's come back to medium style strength to begin with. I'll hit generate. If you'd like to decrease the number of tokens required for the generation, disable the photoreal function. We've got an interesting style effect. Actually, I like it more than the one we've got from the previous generation. And since we are experimenting, let me move the style strength to high and see the difference. All results look good. But I think that medium style strength works best for my project. If you are done with your experiments, download the image, to upload it directly to your Instagram feed or to an editing software like Canva or AdobExpress. If you need to further manipulate the image and add text or other elements. In the Q&A, please let me know what you think about this use case and AC in the next video. 22. Use Case 2: Product Prototyping: One. Welcome back. Let's continue exploring practical use cases for Leonardo AI. The second use case we will discuss is using Leonardo AI for product prototyping and product visualizations. This is useful for pitching ideas to investors conducting market research or gathering feedback. Before committing expensive manufacturing processes, imagine you work as a product designer in a company that specializes in the design and manufacturing of robotic toys. Each toy is equipped with advanced features to assist in learning, sounds, rhythm, motion, and other critical developmental skills. You and your team would like to conduct an ideation workshop to get some inspiration and fresh ideas for a new toy design, you decide to use Leonardo AI software for the job the homepage, open the image creation function. Type in your prompt. This is what I've prepared for this demo. Next, you can experiment with different models that are already available in the catalog. Also, for this use case, you might want to customize your own model based on the image of your previous designs. We will discuss how to do this in the upcoming sections of the course. So please stay tuned. I think I'll stick with lifelike Vision AI model. I'm also fine with all other settings, so let's click Generate. The first generation looks fascinating, but we can go beyond this by including preset styles and elements. Let me add a preset style first. I'll choose three D render and click Generate one more time. But all results look great. And let me also add an element. Let's click View More. Let's add three D sculpt. As for the element strength, I'll keep it unchanged for this iteration, and I'm going to hit generate. Awesome. These images turned out to be very stylized, cartoon like. I really like that the dragons have a friendly appearance with a soft blue body and contrasting yellow accents on their horns, claws, and underbelly. This type of design would work well for children's toys. Lastly, don't forget to experiment with including reference images. I created this holographic three D sphere with Leonardo. Now let's use it as a style reference. Let me choose this image as a style reference image. And let's click Generate to see the changes that the Leonard DI will introduce to the generation. Such a drastic difference between our previous cartoon style dragon and this generation. The images are of a futuristic robotic design. The style was picked up very nicely as we clearly see sleek metallic segments that reflect a variety of colors from the reference image. Remember that when using elements together with the style reference, Leonardo recommends increasing the strength of the elements more than is usually needed. Let's try to do this. So I'll increase it to 1.7 and click Generate one more time. Cool. Of course, some of the elements are quite sharp and are not suitable for kids toys, so they must be changed in the next design version. All right. To sum up this use case, while using tools like Leonardo AI cannot replace using professional three D design software, it can be a great tool for the design team to quickly ideate on possible design variations, develop a product concept, and get feedback from users and colleagues. Please experiment with this use case as well. And if you need prompt suggestions, I leave a PDF file with ten different prompts. You can try to create your first design for a robotic toy. I'll see you in the next video where you will learn how to create prototypes for software products. 23. Use Case 3: UX/UI Design: Everyone. Welcome back. Let's continue exploring Leonardo I's use cases. Apart from brainstorming and developing concepts for physical products, you can rely on Leonardo for ideating interface designs for mobile or web applications. Let me reiterate. I'm talking about an ideation phase. We usually go through when designing new software products or product features, instead of doing sketches on a whiteboard or a paper, you can give Leonardo several prompts with keywords describing what you want to create. And then look for the most unusual unexpected visuals you can take into work. However, you will still need to have a professional design software like Figma. To design high fidelity prototypes, you can test with users. Let's open Leonardo to see the demo. Here is the prompt that I'd like to test. Here, I have information about the nature of the app I'm designing along with a collection of keywords such as Figma, a popular design software, UX, UI design, mobile app design, and others. I also included this phrase, interactive button elements because in my tests, prompts without it often result in just images without any buttons that we would typically expect to see in the app. Don't try to create very long prompts with a lot of details as my tests show that longer prompts often produce worse results compared to those generated with shorter prompts. This prompt you see here is the longest I would use for app design. Let's set the aspect ratio as two to three, since we are designing a mobile app, and in terms of the model selection, I've tested most of the models for this use case and found that these perform the best Albedo, Leonardo quino Leonardo Lighting, as well as Leonardo Diffusion Model. Please remember that Leonardo diffusion model is not available in this list, and to select it, you need to go to the advanced settings and choose the model from the drop down list. I recommend starting with Leonardo ten Excel, as it is cheaper than other models. You can switch to other models later if you are not satisfied with the results from the ten models. I'm fine with all other settings, and let me click Generate. Here are the results. Not all of them look exactly like app designs, but remember that we are currently at the ideation stage. So these are by no means high fidelity prototypes that you can show to users. Another technique you can try is to include the word masterpiece in brackets at the very beginning of the prompt. Using the word masterpiece suggests that the generated image should not only be functional but also visually impressive and artistically significant. Placing it in brackets emphasizes that this term is separate from the functional requirements of the prompt. Okay, let's generate. Another thing you can try is to add the word wireframe after the word app. Wireframe is a basic two dimensional visual representation of a web page, app interface or product layout. This term might help guide Leonardo AI in the right direction. I'll also leave all other settings without change and click Generate. H. I think we've got some interesting prototype ideas, but I would still experiment with choosing other models to see if we can get better results. Let's do this. I'll chose Lanarduqino and heat generate. Nice results, and let's do the same tests for the remaining two models. All right. And the last model that we need to test here is Diffusion Excel. Let's choose it from the advanced settings and click Generate. And it's amazing how the output varies when we change the model. Please let me know what's your favorite iteration out of these four that we've just tested. If you'd like to experiment with this use case, I would advise you to play with the prompt, try different descriptors and combinations, and also test out these four models to see which one work best for the app you are designing. And if you are tested all of these but still want better quality mockups, consider training your own model. This is an advanced technique that we will cover in the subsequent sections of the course. Okay, and that's it for this tutorial Alca in the next one. 24. Taking a deep dive into the world of AI : Hi, everyone, and welcome back to the new section of the course. Imagine stepping back in time and telling Leonardo Da Vincia that one day machines could create art. And now we are very close to making it a reality. So let's dive into how AI is not just learning from us, but also creating with us, challenging our understanding of creativity and innovation. In the upcoming three lectures of this section, let's cover some AI basics, giving you essential knowledge that will enhance your creative journey and deepen your understanding of AI capabilities and potential. We will start with an overview of the AI landscape as of today, speak about how the art generation technology evolved over time and why it recently became so popular with the explosion of new software and tools. And of course, we will discuss how the art generators work and how you can go from a simple text description to something like this. This section is optional. So if you'd like to focus on the practical application of art generation technology, feel free to skip it for now and come back to it later whenever you are ready to learn more about AI. And for those of you who have decided to follow along, I'll see you in the next video. 25. AI landscape of today: Everyone, welcome back. In this lecture, we will go through an overview of the AI landscape as of today. First of all, let's define what AI is. In simple terms, AI is the ability of machines to learn, understand, reason, and interact in ways similar to us humans. This allows machines to solve new sets of problems they could not before. For example, AI powers voice assistants like Syria, recommends movies on Netflix and help doctors diagnose diseases. AI encompasses a range of technologies from simple automated rules in everyday gadgets to advanced systems that learn and adopt while AI can perform specific tasks at or above the human level. At the moment of recording this video, it does not possess general intelligence or consciousness. Recently EI has also made significant progress in creative fields, generating art, music, and literature. That's exactly what we are going to explore in this course. Now that you understand what AI is, let's discuss how machines actually learn. At its core, machine learning, a key component of AI, involves teaching computers to recognize patterns and make decisions based on data. This process is somewhat similar to how humans learn from experience. But instead of learning from life experiences, machines learn from data. Machines learn in different ways, mainly categorized into three types, supervised learning unsupervised learning and reinforcement learning. These are what we call the foundational learning methodologies. Each of these methodologies has its own approach to learning and is used for different kinds of tasks. Supervised learning involves training AI models on labeled data. Labels are identifiers associated with input data. For example, they can be textual. In a dataset of animal potas. Each pota input would be labeled with the name of the animal output like cat, dog, et cetera. Another example is numerical labels that can be used to predict house prices based on features. Supervised learning essential for applications where the model learns to predict outcomes based on provided examples. This includes speech recognition, image classification, and expert systems. AI systems that mimic the decision making abilities of a human expert in a specific domain. Unsupervised learning focuses on finding patterns or structures in unlabeled data. In other words, it discovers the underlying patterns in the data without explicit guidance. The unsupervised learning is pivotal in domains like recommender systems, systems that predict user preferences and suggest relevant items accordingly. It is also used in certain aspects of computer vision that focuses on enabling machines to interpret and respond to visual information from the surrounding environment. Third methodology is reinforcement learning. It focuses on training models to make decisions through trial and error, receiving feedback from the environment and learning optimal actions through rewards. It's key in robotics, autonomous vehicles, and some planning and scheduling tasks like resource management and automated scheduling systems. Please note that most application areas rely on a combination of different learning methodologies to leverage the strength of each. This approach often gets better performance and more robust solutions. For instance, many modern recommender systems integrate all three methodologies to leverage their strength. Supervised learning provides accuracy based on historical data, like predicting and recommending new movies or products a user might like based on historical data with user preferences or ratings. On the other hand, unsupervised learning offers insights into users which might not be apparent through ratings alone. Clustering algorithms, a type of unsupervised learning technique that organizes data into clusters or groups based on similarities, might find that certain groups of users tend to watch similar genres of movies even without explicit ratings and recommend movies based on these clusters. And finally, in case you want the recommendation engine to be dynamic and adopt the recommendations based on how users interact with different content. For example, by browsing, watching trailers, selecting and watching movies, reinforcement learning comes into play. This system will learn by interacting with users over time and adjusts its recommendations based on user engagement and feedback. All right, our overview of the AI application areas won't be complete without the other two that also leverage all three foundational learning methodologies. These application areas are natural language processing or NLP and generative AI. ALP implies understanding, interpreting and generating human language, and is used in such applications as language translation, sentiment analysis, chat booard and voice assistance. And finally, generative AI, the term that has become extremely popular in 2023 and that you probably have heard of before. It is an umbrella term that includes various techniques focused on creating new original content that never existed before that mimics or inspired by real world examples, AI art generation. Something that we will be doing in this course, specifically refers to the use of generative AI techniques to create artworks. It is a niche within generative AI focusing on visual creativity. AI models in art generation learn styles, patterns, and artistic elements from existing artworks using foundational learning methodologies, and then use this knowledge to generate new images, paintings, or visual content. In the following lecture, we will learn more about how the AIR generation technology evolved over time and why it was taken by storm recently. But before we proceed, let's sum up this lecture. AI is the ability of machines to learn, understand, reason, and interact in ways similar to us, humans. A key component of AI machine learning involves teaching computers to recognize patterns and make decisions based on data. Machines learn in different ways, mainly categorized into three types or foundational learning methodologies, supervised, unsupervised, and reinforcement learning. Supervised learning teaches AI with labeled data. Unsupervised learning finds data patterns without guidance, and reinforcement learning involves learning via feedback. Most application areas rely on a combination of these learning methodologies to leverage the strength of each. Generative AI is an umbrella term that includes various techniques focused on creating new content that never existed before inspired by real world examples. All right. That's it for the lecture and I'll see in the next video. 26. Evolution of AI art generation technologies: Everyone, and welcome back to the series of lectures where we cover AI fundamentals. In this lecture, we will find out how AIR generation technology came from being able to do this to this and even this. Let's begin. Generative AI research traces its history back to the 1960s. However, generative AI began to develop into something similar to its current form in 2006. The first significant paper in the field by Jeffrey Hinton and his co authors titled fast Learning Algorithms for Deep Belief Nuts. But the first major breakthrough in the field of image generation happened back in 2014 with the introduction of a novel framework called Generative Adversarial Networks or Gans prior to introducing Gans, AIs focus in the visual domain. Was predominantly on image recognition and classification. Ganz shifted this focus towards the generation of entirely new images. Initially, they weren't used for turning text into images, but rather for creating realistic images from random chaotic visuals. The Gans then gradually transformed these initial chaotic visuals into coherent and realistic images. Gans brought a significant shift in how machines could create visually detailed and realistic images, laying the groundwork for more sophisticated text to image applications. Let's try to go a bit deeper and see how the model works. In a natural, the framework implied two neural networks, a generator, and a discriminator working against each other, leading to improved quality of generated images. This description is quite technical, so let's simplify it for a bit. Imagine two people, let's call them the artist and the detective. They play a game to trick each other but in a fun and creative way. The artist loves to draw pictures, but instead of drawing from real things, the artist tries to create drawings that look like they could be real, even though they are completely made up. Think of the artist trying to draw a unicorn that looks like it could exist in real life. The detective is really good at figuring out what's real and what's not. So when the artist shows a drawing, the detective tries to guess if it is a real thing or just a make believe drawing by the artist. The artist keeps making new drawings, trying to make them looks as real as possible, and the detective keeps trying to guess correctly. As they keep playing, both the artist and the detective get really good at their jobs. In the world of computers, the artist is like one part of the game that creates things, a generator, and the detective is the other part that checks if they are good enough or not a discriminator. By working together and challenging each other, they end up creating really amazing things that can sometimes trek even humans into thinking they are real. Following the breakthrough of generative adversarial networks in 2014, the development of text to image models has seen several key milestones. In 2017, researchers at Google introduced transformer models that revolutionized the field of machine learning. Particularly in tasks involving natural language processing. Over time, researchers discovered that transformer models could be applied to visual data as well. One of the key features of transformer models is the ability to process entire sequence of data simultaneously, unlike traditional models that process data point by point. This feature enables transformers to understand the context and relationships within the data more effectively, making them particularly powerful for tasks in natural language processing and image processing. This became possible due to the attention mechanism, which is a cornerstone of transformer models. This mechanism allows the model to focus on different parts of the input data prompt and understand how each part relates to the others. You can visualize the transformer model as a super smart robot that can look at all the pieces of a huge Dixopuzzle at once and figure out how they fit together really quickly. It is especially good at solving puzzles with words or pictures. Understanding which pieces are important and how they all connect to make up the whole picture. So if you tell a story or show it a bunch of drawings, it can quickly make up a new story or picture that fits everything together perfectly. Since its introduction, the use of transformer models and image generation was largely experimental. Researchers were exploring how to adapt these models, originally designed for text. To handle visual data. Everything has changed in 2021, when a major tech player company called Open AI has introduced Dali, an AI system that can create realistic images and art from a description in natural language. Dali represented a significant leap in the capabilities of AI creative tasks. I demonstrated an unprecedented level of proficiency in generating diverse, complex and contextually accurate images from textual descriptions. This showcased the potential of AI in creative domains, far beyond what was previously thought possible. Seeing the capabilities of Dali developers and tech companies were inspired to explore similar technologies. This led to a wave in innovation and development in generator software. As more entities thought to leverage the underlying technology for various applications. The most prominent art generator tools that have emerged since the introduction of Dali include mid journey, stable diffusion, open art, Firefly, Leonard the AI, and others. Around the same time when Open AI announced Dali, another major breakthrough in the image generation field happened with the introduction of diffusion models. These models showed impressive results in generating high quality, detailed and coherent images, often rivaling or surpassing the quality of images generated by Gans. Diffusion models unique process involves starting with an image that is entirely noise. Think of a TV static screen. Over multiple iterations, this noise is slowly reduced with each step bringing the image closer to a realistic and coherent final picture. We will speak more on how the diffusion model work in the next lecture. After watching the lecture, you will know exactly how the generator software works and how it translates a text prompt into the stunning images. But before we go there, let's sum up what we've learned in this lesson. Generative AI's history dates back to the 1960s, evolving significantly with Jeffrey Hinton's 2006 paper. The introduction of Gans in 2014 marked a major advancement, shifting the focus from image recognition to generation. Gans use a generator and discriminator network to improve image quality. Similar to a game between an artist and detective, transformer models introduced in 2017 revolutionized the field of machine learning and were applied to both language and visual data. Dali, an AI system that can create realistic images from textual descriptions showcase the potential of AI in creative domains, far beyond what was previously thought possible diffusion models, another breakthrough in the image generation field, generate images by refining them from noise to detailed pictures. All right, that's it for the video, and I'll see you in the next. 27. How AI generates art: Everyone, welcome back. In this lecture, let's explore how AIR generation actually works. What happens after you enter a text prompt and hit generate, and if and how can you influence the R generation process? Let's begin. The AR generation process consists of the following four steps. Let's talk about each step in more detail. Step one, data gathering and preprocessing. The process begins with collecting a large dataset of images. Here we are talking about really large datasets comprising hundreds of millions of images, as well as the textual descriptions. The larger the dataset and the more diversities, like having images with a wide range of subjects, various artistic styles, lighting conditions, and compositions. The more varied and nuanced the learning can be and so as the final generated images, it's like giving an artist a broader range of experiences to draw inspiration from. Once the data is gathered, the images are then preprocessed to ensure consistency in the data fat into the model so that it can learn faster and more efficiently. Imagine if some images were blurry or oddly colored. Without preprocessing to normalize this, the model might learn incorrect or inconsistent visual patterns like mistaking a blurred image for a style. Once the data is collected and preprocessed, we are ready to go ahead with the step two of the process model training and latent space formation. While it is technically feasible for an AIR generator to use multiple types of models like Gans, transformers or diffusion models, we've covered earlier. Such a system would be complex to implement. Most current tools tend to focus on a single model type. For example, Dali primarily uses a transformer model, and Firefly is a diffusion model. Whichever model is chosen, it undergoes a training process where it learns to interpret text prompts and generate corresponding images. During training, as the model learns from the data, it creates a multidimensional latent space. This space abstractly represents the various features of the images like style, color, or content in a lower dimensional format. It's getting technical. So let's take an example. Imagine we are training a model on a dataset of different animal pictures. During training, the model develops a multidimensional ten space. Let's simplify this concept by imagining it as a big invisible map. Each dimension in this space represents different features of the animals. Think of one dimension for color like brown, white, black, another for size, small, medium, large, one for type, mammal, bird, reptile, and so on. Every point in this space is a combination of these features. One point might represent a small brown mammal like a squirrel while another point could be a large, white bird, like a swan. It's hard to visualize many dimensions. So let's simplify it further. Picture a two dimensional grid. The horizontal axis represents size. Left is small, right is large, and the vertical axis represents color. Bottom is dark and top is light. A point on this grid shows an animal's size and color. When the AI wants to create an image of a large, dark colored animal, it moves to a point on the grid that is towards the top right corner, large and dark. This point in the ten space corresponds to the features of the animal it will generate. The actual ten space is far more complex with many more dimensions than just two or three, often in the range of hundreds or thousands. It's not something you can see with your eyes. It's more like a mathematical concept stored inside the machine AI is running on. However, I asked to visualize the multidimensional latent space for me, and that's the picture that I got. Not bad. The AI learns to navigate this complex space during training. By moving around in this space, it can generate a wide variety of images. In our example, animal images, each with different combinations of features. All right. After the second step is done, we are ready to go ahead with the fun part. Step three, generating art from a text prompt. The AI model uses your text prompt to navigate the latent space, finding points that correspond to the desired features or styles. Every model has its own unique mechanism for navigating the latent space and generating images. Let's see how the diffusion model works when it comes to generating images from text prompts. The model starts with a canvas of pure noise, a random arrangement of pixels. When given a prompt like a cat on a sofa, the model uses this input as a guide setting the direction for the transformation process. The model then iteratively refines the canvas by navigating different regions or cardinates within the latent space and introducing and sharpening features relevant to the input prompt. This is what is called reverse diffusion or removing the noise from the initial canvas. As the noise diminishes, features start to emerge. Based on the prompt and its learned knowledge, the model begins to introduce elements like the shape of a cat, the texture of a sofa, so on and so forth. Stage represents back and forth between the noisy and less noisy states where the model decides which features align with the given prompt. The final image is a coherent and realistic depiction of a cat on a sofa. This image represents a specific combination of features encoded in the laden space, the size, color, and pose of the cat, the style and color of the sofa, all influenced by the initial prompt and the models straining. Now you may have a very reasonable question. But how come the model decides on what's the size of the cat, its color or pose? These details come from the input prompt. In case you have a simple prompt like a cat on a sofa, the model might default the final image to more commonly seen or average representations of cats based on its training. So by adding more details to the prompt, for instance, including such descriptors as Cute for a cat and antique for a sofa, you are pushing the model to navigate through more specific regions of the latent space that correspond to these specific attributes. Cute might relate to certain aesthetic features of the cat while antique could involve particular styles or patterns associated with sofas. As a result, the model generates an image that not only includes a cat and a sofa, but also reflects the specified characteristics of being cute and antique. This leads to a more nuanced and contextually rich image. Of course, there is also an element of randomness in how the model navigates the latent space, leading to creative and varied outputs. So using the same prompt might give you different images each time because of the random way the model navigates its latent space. We are almost at the end of image generation process from the text proms. The final step includes post processing when we can enhance the image by, for example, adding text, adjusting contrast, or even adjusting parts of the composition. The number of steps of how you can alter the image depends on your creativity and the final result you want to get. Needless to say the post processing step is optional. You can use the image generated from AI generator software without any modifications. If you like it as it is. Cool. That's it for the lecture, and as always, let's recap on what we've just learned. AI R generation process consists of four steps. The process begins with collecting and preprocessing a vast dataset of images and their descriptions to teach the AI. The next step includes model training and latent space formation where this space abstractly represents various features of the image, such as style, color, and others. Step three involves generating art from a text prompt with the EI using prompts to navigate the latent space and generate images. Adding specific descriptors to prompts guides the EI to produce more detailed and contextually rich images. The final and optional step involves post processing activities, allowing further customization of the AI generated images. All right, call ACA in the next video. 28. Getting started with Motion: Everybody. Welcome to the new section of the course. Here we are going to take the next step in working with generated images and bring them to life using Leonardo As motion module. With the platforms easy to use interface, creating short form video clips is a seamless and intuitive process. All you have to do is choose an image, either one generated with Leonardo or one from your device. The latter option is available if you work on one of the premium plans. To generate the video, you don't need to come up with the prom description. Just set a motion strength to define how much movement will be added to the video and hit the generate button. But that's enough of the intro. Let's dive into the demo to see everything in action. 29. Let's create your first video with Leonardo.Ai!: Homepage, click on Motion under bring your ideas to live. Alternatively, select motion from the left side toolbar. On the next pop up screen, click on select an Image. Next, choose upload an image to get started in case if you'd like to choose an image from your local drive. As discussed, this option is available for paid subscribers only. You can also choose an image from your own generations or those that you saved into your collections, community feed images, or the images of your follower feed. Let's begin by selecting an image that we've generated before. Let's select this image as it contains several moving elements like these cars in the background, as well as a woman in the foreground. I expect to see the woman moving while she crosses the road. The cars on the other hand should stay still. It will be very interesting to see how the animation works. I'll click Confirm. We can also modify the motion strengths to define how much movement will be added. Let's leave it as this. The motion will be private, not available to other Leonardo users. This option is available if you use Leonardo's paid plans. All good. Let me click Generate. Leonardo will take some time to generate the motion to check your video, go to your library. And if it's not here yet, it means that the Leonardo is still generating it as we speak. You can go to image creation section and you will see that the image generation is in progress here. Let's wait for a bit. How clip is ready. Let's click on it to enlarge it. We see that it is far from perfect, most likely because we selected a person we wanted to move. On the other hand, notice how Leonardo identified which parts of the image should be animated. Let me show you another example that I got from this image. It is less complicated than the previous one as there are no people or characters involved. However, please notice how nicely Leonardo picked up on the snowflakes movement, creating this coisy winter atmosphere. I also like the pathway movement as if we are walking through this winter forest and exploring the surroundings. Now it is your turn to try this feature in action and animate your creations and I'll see you in the next video. 30. Introducing Advanced Techniques: Training your own AI model: Hello, and welcome to the new section of the course. If you followed along with the previous lectures, you should already be familiar with how to work with Leonardo AI models. How each model is different from the others, and most likely you have some preferred models that you choose for your projects most of the time. However, there might be cases when predefined models cannot suit your needs anymore. For instance, when you would like to keep creating images of a certain style or theme or create images with consistent character designs of your choice, the existing models won't work here as they weren't trained on images of that specific style or with that specific character design that you want to create. What you can do here is train your own model and then use it instead of the predefined models available on the platform. Let me warn you up front that training a model is a premium feature. So if you are on a free plan right now, consider upgrading your plan to follow along with the tutorials. In the upcoming lectures, we will go through every step of designing a model. We will first create a dataset of images needed to train the model and learn what images will and will not be suitable for model training. Next, we will train the model and try it in action. Finally, we will refine the model in case we see any imperfections with the generated images and would like to improve the original dataset. For the tutorials in this section, I will pretend that I'm working in an interior design form, and I'd like to create a series of posts for the company's Instagram account. Featuring Memphis style interior design. The Memphis style, a design movement that emerged in the 1980s, is known for its eclectic mix of geometric shapes, bold colors, and unconventional patterns. There is no model available in Leonardo AI's collection of the fine tuned models that could accurately generate images in this style. So I decided to train my own AI model. So if you are ready to follow along with me, let's get started. 31. Creating a Dataset: One, welcome to the first tutorial of this section. Let's begin by creating a dataset of images that we will use to train the model. Here are the images I've selected that represent Memphis style interior design. I'd like to create designs for a living room with a sofa in the center. A coffee table in front of the sofa and armchairs on the side of the sofa. That's why I created images that depict both the style and type of design. I'd like the new model to replicate. I have to warn you that creating the right dataset is the most challenging task. So let's go over some key rules to keep in mind when selecting the images. First of all, select high quality watermark free, non blurry images. I recommend using Adobe Stock for this. You can get a 30 day free trial that gives you access to ten images, which will be enough to train the model. That's how I created my dataset. Here is what could happen if you select low quality images. Here are the images that I downloaded from Pinterest. You can see that even though they depict the Memphis style quite nicely, their quality is not the best, and many of them look blur rim. And here is what I got as the final output after I trend the model with these images. The resulting images are quite blurry, and the details of the geometric patterns on the walls, the shapes of the furniture, and the textures are all indistinct and hazy. That's clearly not the effect I'm aiming for. Okay, now let's move on to the second recommendation for selecting images for your dataset, and for consistency in terms of style, format, and aspect ratio, while still introducing variations within these constraints. Finding the right balance between consistency and variation is something you usually achieve through multiple iterations. Training the model, checking the output, and then making changes to your dataset before re training the model. For instance, this image won't work because I like to create design with the sofa as the central object of the image with the coffee table in the foreground and armchairs on the sides. On the other hand, this image won't work either since here isn't enough space in front of the sofa and the objects on the right and left sides of the sofa aren't the ones I'd like to see in my final images. In my dataset, I've tried to collect diverse pictures of my chosen design. For instance, you will see a range of colors and variations of designs for the sofas, coffee tables, and armchairs. As I mentioned earlier, you need to collect images with the same aspect ratio. If you need to resize your original images, I recommend using Canva. Which is available for free. Create a document of your target size, upload your images. And then drag and drop them to the Canvas. I like this method because it allows you to quickly spot images that won't work well for the model, like in this example, the third recommendation I'd like to cover here regarding image selection is that your dataset should include up to 40 high quality images. If you include more than that, your trained model might end up recreating the training dataset instead of being able to generate new variations. For this demo, I'm using a small dataset of up to six images, and as you can see in the upcoming tutorials, I was able to achieve decent results with this small dataset. Right. As soon as you have your first set of images ready, go to the Leonardo Training and dataset module. Here, click on New dataset. Type in the name and description. Click on Create dataset. Here, click on Upload Images and select images that you prepared for your model training. And we are all set to start training our first model for this, let's jump into the next tutorial. I'll see you there. 32. Training the model and testing it in action: One, welcome to the next tutorial in this section. Now that your dataset is ready, let's train our model from training and datasets module hover over your dataset and click Start Training. Fill in metadata for your model to help with categorization and retrieval. These include elements such as model description, category, and prompt instance. You can choose a model category from the list. I think environments works best for my project, and as for the prompt instance, think of it as a simple way to guide the model to properly utilize its training data framework. For instance, for a watercolor style model, it might be something like a watercolor painting of in my example, I'll write living room design. I'll leave all other settings such as training, resolution, as well as base model without any changes. And I think I'm ready to hit start training. The training process usually takes 30-2 hours depending on your dataset. For small datasets like mine, it's even faster. Let's wait. It took about 2 minutes to train the model. Now, let's test the model in action. For this, go to image creation module. Here is something that you need to be aware of. Your trained models are only available in the legacy mode of the image creation tool. Legacy mode is an old interface that existed before the Leonardo team introduced the updated version, which you see on the screen right now. Let's activate the legacy mode and wait for a few seconds. And here we are. This interface can be a bit intimidating for you as we didn't cover it in the previous tutorials, but no worries. I'll explain everything that you need to know to be able to test your own model. The first thing that you need to do is type in a prompt that you'd like to test. This is the prompt that I will be using for the first generation. Next, let's select the model that we've just trained. Open the list of models, go to select other models. And from here, open the tab your models. Click on view and then select generate with this model. Our new model has been selected, and now let's see if we want to change any other settings here. You can choose preset in case if you'd like to make any alteration to your original image, you can choose elements. I recommend to leave both of the settings without changes, at least for the first iteration, so that you can see how your new model performs without any additional elements and customization. As always, you can choose a number of images that Leonardo AI will generate. Here is the photoreal function that you can activate in case you are generating photorealistic images. Again, for the first iteration, just switch it off. Alchemy mode is an old version of the quality generation mode that we already covered in the new interface. It was selected by default, and for this demo, I'll leave it without any changes. Here are some other settings that you can modify as well. Here we have transparency Togle that I won't be activating for my project. If we want our images to be publicly available to all other users, switch this indicator on and let's check other settings here. You can choose from a variety of input dimensions, or you can select the aspect ratio for your newly generated images from this list. I leave these settings without any changes, and I think I'm ready to hit generate. Our images turned out to be really nice, especially considering that I used a very small dataset of images and that this is our first iteration. Personally, I feel like I would introduce more variability in terms of the colors to this dataset, and you can do this quite easily by refining your current dataset and retraining your model. This is what we are going to cover in the next tutorial. I'll see it there. 33. Refining your model: Everyone, welcome back. If the results you see from testing your model are not satisfactory or you want to make some improvements, you can retrain a new model by going to training and datasets. Choosing your dataset and selecting died dataset. You will be able to delete and replace images, or you can add additional images to your existing dataset. That is what I'm going to do. Here is an image that I'd like to add to my existing dataset to add more color variability into it. Unfortunately, it's not possible to update an existing model that has already been trained due to technical limitations. This means that every time a dataset is modified, a new model must be trained to reflect the changes made. Let's train a new model. For this, I'll click on Start training. And here, I need to select model description category and prompt instance. Let's do this. So you can check the job status by going to the job status tamp and clicking on refresh. Here we see the second job, and it is currently in progress. So let's wait for a few minutes. The training process has been completed. This time, it took me a bit longer time to do this. So the time required to train the model can vary. But that's fine. Now we are fully ready to test our second model. Let's do this by going to the image creation function. We are still at the legacy mode interface, which is perfectly fine for our project. Let's choose the new model by going to select other model, your models, and here we have the newly generated model. I'll click on view and then select Generate with this model. For the purpose of our experiment, I won't be changing any other settings that we had for our first iteration here and let's just click on Generate. And here are our results. We see that the color schemer on these new images is definitely different from the one we got from the first generation. I really like these pastel colors, and I think they work perfect for the project that I envision. However, you can continue experimenting with modifying your dataset by deleting some images and adding a new one and then training a new model. Okay, that's it for this tutorial. Of course, now I encourage you to try out training your own model yourself and please let me know in the Q&A section for this video. What do you think about this functionality and if you find it useful? I'll see you in the upcoming videos. 34. Introducing Advanced Techniques: Generating images from Drawings: Everyone. Welcome back to the new section of the course. If you have experimented with image generation from text instructions or prompts and feel like instructing AI with just text descriptions is not enough to convey your ideas or vision for the final output. I have great news for you. With Leonardo, you can now create drawings that will be converted into artworks almost instantaneously providing you with an entirely new way of interacting with AI. In this section of the course, we will cover the real time Canvas module of Leonardo AI, which allows you to do just that. With the real time canvas, you can easily ideate visuals by simple sketching, even if you don't consider yourself a professional designer or artist. You will be amazed at how simple sketches are turned by Leonardo into detailed images, and when coupled with your prom description, they will come as close to your original ideas as possible. I'm really excited to introduce you to this module, and let's begin. 35. Getting started with Realtime Canvas: One, welcome back to our third tutorial, where we are going to talk about the real time Canvas. Let's begin by opening the canvas. It is available from the AI tools on the left hand side tool bar. Let's start with the prompt wooden cabin. And we already see that our first image has been generated instantly. But let's see how far we can improve it with the drawing tools available here in the left hand side bar. The first tool you are going to work with very often is a brush. Let's begin by adjusting the brush size as well as the color. For my work, I'll choose the brown one, then I'll start drawing the wooden cabin. Actually, let me reduce the brush size first. It may be quite challenging to draw a straight line, so you can press, shift, and continue drawing. Now we have the perfect straight line. You see that Leonardo adjusts the output image every time I add a new line to the canvas. If you like to just the position of a line, click on this icon. Select a line that you'd like to move and then move it or either reduce it or enlarge it. We have something interesting on the right hand side of our canvas, but I think that our picture is very dark. Let's change that. I'll go to color. Here I'll select background. Let's choose something from the blue color scheme, something like this. Wow, the difference is incredible. I can hardly imagine that this super simple sketch can be transformed into this stunning fantasy like image. Here is the setting that helps with the creative transformation. Creativity strength. If we lower it to a minimum, we pretty much get the same picture as we see on the canvas. On the other hand, if we increase it to its maximum, we will get more creative EI interpretation of our drawing and the prompt. Let me lower the creativity strength for a bit. Try to play with the setting to find the best balance for you. Okay, another feature that affects the final output is style presets. You can quickly apply a specific style to your project without the need to add any complex prompts. Let's try several presets. I like the environments preset best for my project, so I'll keep it. Okay, let's finish with the settings on the right hand side of the bottom tool bar. Here, you can choose between the real time mode, which is a premium feature and interactive mode. When real time is selected, the real time Canvas turns each brush stroke into detail in near real time. With interactive mode on the real time Canvas waits for your drawing actions to complete before transforming your sketches into detailed artwork. Let's switch to interactive mode to see how it works. Okay, let me switch back to real time and continue editing the sketch. If you'd like to raise the line, click on the eraser and make the changes. A Please note that you can change the transparency of every line that you draw. For example, in case if you'd like to add a bit of light on this side of the roof, just move the slider to the left, switch color to whitish, and then draw a line here. You see that the line is transparent. This picture looks like a snow to me, so I would probably increase the transparency even more. Or maybe let's change the color for these lines to the yellow spectrum. Now we have a bit of a lighting on the right hand side of the roof. However, it cannot be too bright as the surroundings here are quite dark and we see that the lights here is lid so the bright light over here would create an artificial atmosphere. Let me remove those lines. Return the slider to its maximum and draw a perfect white solid line as I'd like to have some snow on the right hand side of the roof. Yeah, something like this looks great. And let's also add a few more changes. Okay, we've got pretty interesting result on the output image side. We can continue plan with the creativity strength to see if we can improve the results. Okay, let's leave it like this. If you want to add more variations for the output, go to the advanced settings by clicking this icon here, disable the fixed SID option and then click on random SED to generate a new image with your current prompt and settings. Also adjust the guidance. This controls how closely your artwork aligns with the text prompt, set higher values for greater adherence. Of course, you can continue including more details into your prompt to improve the final output. And once you are happy with your image, you can download it to your local drive to use a or continue editing it later on. Before you download, consider refining the image by clicking on Instant refine. This feature refines the output within the canvas and increases its resolution to 1024 by 1024 pixels. This process has no token costs. Once the refining process is completed, click on Download to save the refined image to your local drive. Another thing you can do is upscale your image. Click on the settings on the right hand side of the upscale image button. And here, choose a refiners strength as well as smooth mode to improve the image coherence. Then click on Upscale as you remember, we talked about image refinement and upscaling, as well as these settings in the lecture on how to enhance your image. Please rewatch it one more time if you need refresher. Once the upscaling process is completed, click on Download to save the adjusted image into your local drive. Okay, that's it for this tutorial. In the next video, we are going to talk about more opportunities to improve your output image. We will cover some pretty cool features, so please don't miss this lecture and I'll see you there. 36. Enhancing your drawings with Output to Input feature: One, welcome back to the tutorial on real time Canvas. As promised in the previous tutorial, let's see what else you can do to improve this image. With real time Canvas, you can define the outputs further by using the output input feature. You see that when I click on this button, our output image has been copied to the canvas and now you can make changes to it by sketching on top of it. I think this is a very cool feature as it gives you pretty much endless possibilities to improve the output. Let me draw something here. Leonardo doesn't quite understand what I like to see in the output image. Let's help it a bit with adjusting my prompt. Okay, perfect. That's exactly what I was looking to achieve. Let me also switch back to using a fixed seat to reduce the variation of my output. Okay, let me also reduce the creativity strengths for a bit. This picture looks nice. Let's see if we can get anything better than this. No, that's too bright. You know, I think I'll definitely increase the creativity strength. Oh, this one is nice. It's quite a time consuming process to find the right balance between the creativity strength parameter, as well as your final output. I think I'm fine with this picture. Again, remember that you can also continue playing with these presets. Let's try fantasy art one more time. I actually like photography, so let's keep it for now, let's do output to input several more times to continue experimenting with the image. You can do this exercise of copying the output image to the canvas as many times as you like until you get the result that you really like and want to keep for further editing. Now, another interesting feature you can explore to enhance your output is generating an additional image with the prompt and adding it to your current image as a second layer. Click on Generate icon. Next, type in your prompt here. If you don't like the first picture, you can click Regenerate as many times as you need to get your perfect image. I think I like this one. Let's click Done. The next thing you need to do is to remove a background from this image. Click remove background. Icon. And here we go. Now we need to find the right place for the snowman in front of this magical wooden cabin. I think I'll adjust the prompt as well. We've started getting some interesting results here on the right. You pretty much play with the position of your new image until you find the right spot for it and satisfy with the output on the right hand side of the canvas. A when you are happy with the image, as always either download it as this or first do instant refining for the image or upscale it, and then save it to your local drive. The third thing you can do to improve your generated image is instead of generating a new image layer from scratch, you can add one from your local drive. This can be external image that you'd like to add to your project. For instance, let me choose this one from my local drive. Let's remove the background as well. Again, you should play with the position of the moon and you prompt to get the best final outcome. These results look quite interesting. I think I'm going to keep it. So let me save this image. But before I do this, I'll do instant refining. I can continue playing with the position of the moon here, as well as with other settings that we have in the canvas. Okay, great. That's it for this tutorial and Alca in the next one, where we continue exploring the real time canvas. 37. Combining several of your existing and newly generated images in Realtime Canvas: One. Welcome back. In this quick tutorial, I'd like to show you how you can combine several of your existing images into the canvas to create a creative and unique final image. Let's begin by removing everything from this canvas. Make sure that you saved the output images into your local drive before you do this. I'll also remove the prompt. And let me add image from my local drive. So here is an image that I took from the Community feed and I'd like to use it as a background for my new image. Let's change the background color to black, I'd like to use these skyscrapers as a background for my new image, and let's generate an image that will be on the foreground for this, I'll click on Generate Okay, let's remove the background. Okay, it looked quite good. And let's copy this image. I want to duplicate it. We see that something is happening on the right hand side of the canvas, but it's not even close to what I'd like to create. Now let's add a prompt. Okay, the first generation look nice, but I don't want anyone to be here. So let me actually choose a different preset. Let's try environment. Yeah, this one is much better. We can also try fantasy art. Perhaps you have your favorite preset, so you can start with that one. Let's also try photography. Okay, I think environment is my favorite so far, so I'll return back to it. As always, let's enhance the creativity strengths. I usually try to increase it up to its maximum to see what we get. Yeah, I think this is a bit too much, so I'll start lowering the creativity strength up until the moment when I find the right balance. Okay, this result looks best for me, but that's not all. Let's also add my second image that I saved to my local drive. Actually, I already uploaded it here for our previous demo. So let's use this moon for this project as well. I'll click Confirm. As always, let's remove the background. Okay, I'll just the size of the moon and place it here. It's already quite nice looking picture on the right hand side, but let's also add a prompt. Looks very nice and quite surrealistic. Okay, let's stop here for the purpose of our demo. Frankly, I can spend endless time playing with the position of my images as well as with the settings that we have here in the real time Canvas. But I think that by now, you have a pretty good idea of how much you can do with the real time Canvas module. By combining your existing photos, the photos you generated with Leonard Dui into a new creative, unique artwork. Now, as always, I encourage you to start your own experiments and please feel free to share your work in the Q&A section for this video. And that's it for this video. Bye bye.