Master AI Video Generation using ComfyUI and Minimax H3 in your computer

1. Introduction: What if you could build a full AI video portfolio using CFIUI for free on your computer. Hi. I'm Karen your instructor for this course. In this course, you are going to learn how to create videos, images within your computer. That is absolutely free. What you will learn in this class, we will use Minimax S three inside CFUI. You will start with the interface and basic workflow, so nothing feels confusing. We will use skills free AIchatPot, that is chat GPT, Google Gemini, and Cloud within free structure to understand how prompting work. So you will never stuck at any step. You will also learn the basics of Omnimodal video generation. So we will some theory part also. Then comes the part most courses escape. I will show you how to take a low quality generation, upscale that very same video and generate it ready to upload in high quality anywhere. And along the way, we will create several projects. So by the end, you will have an AI video portfolio. You can showcase, share your own clips in the project tabs. I can't wait to see it. So what's the minimum requirement? A PC with an NVDA graphic card with at least 8 gigabytes of graphic VM and 16 GB of RAM, 50 gigabytes of free SSD space. Absolutely no coding required. This class is for video editors, designers, content creators, and complete beginners who want to start with AI video. You are overwhelmed by all the AI tools that are available on the Internet that cost you so much and you want to learn them, but do not want to invest in them until you learn them, right? So I recommend you first, learn how the video generation work, and then you can start using those expensive tools for yourself or for your company. Know how painful it is to create videos and images while burning your money. There are so many AI tools on the Internet that are available, but they are very costly right now. So we are going to learn CPI in this course. So what are you waiting for start your journey. Let's get started. 2. What is minimax h3: Before we start building any complex workflow and rendering videos, it is really important that you understand what exactly it is you are installing on your computer. What is Minimax H three? Minimax H three is a brand new open source AI model created by a company called Halo AI. S three stands for third generation of their technology. It is a massive model built over 33 billion parameters, which is just a technical way of saying that it is very, very big and very, very smart brain. But what makes Minimax a three so special? Why is everyone in the AI community talking about it? It comes down to two major breakthroughs. Omnimodal architecture and native AI audio. Let's break those down simply Omnimodal architecture. In the past, AI models were usually specialized. You had one AI that only read text to make an image. You had a completely different AI that look an image and made it move, and you had another completely separate AI that generated voice. Minimax SE is Omnimdal. This means it's understand everything all at once. In a single generation, you can feed it text instruction, multiple reference images, a short video clip. And an audio file and it will look all of them together to understand exactly what you want. It is a true all in one general purpose model, native stereo audio. It is a feature that changes everything. Traditionally, AI video models only generated silent moving pictures. If you want sound, you had to take that silent video into an editing program like Premiere Pro and manually add music sound effects or voice. Minimax has three generated the video and the stereo audio at the exact same time in a single pass. If you prompted to generate a video of a busy city state in the rain, it doesn't just show you the rain. It generates the sound of the raindrop. This is the prompt, a busy city state in the rain. Let's play this one. It is a two second video that I have created. What you notice, video and rain sound and the car's sound. It generates the sound of the raindrops hitting the pavement and the cars driving by, right? Perfectly sync to the video. If you give it a picture of a character and an audiophile of a person talking, it will animate the character's mouth to perfectly live sing the word. No, now we are talking about output. When you run Minimax as three, the final file you get is the standard MP four file. It can generate clips up to 50 seconds long running at a smooth 24 prime per second with resolutions up to two K high definition. And the best part because it is openwight, it is completely free to download and run locally in your own hardware without paying any subscription fees. Now that you know how incredibly powerful this model is, let's move to the next lecture. I will see you there. 3. Minimax Video Model system compatibility: In this lecture, we need to talk about your computer. Minimax a three is an incredibly powerful AI model. It can understand video images, text, and audio all at once because it is so smart, it is also very heavy. Before we start downloading anything, we need to make sure your PC has the right hardware. Let's break down three main things your computer needs. Your graphic card, your system RAM, and your storage. Number one, the graphic card or GPU or we can say VRAM when you generate AI videos locally on your computer, your graphics card does almost all the heavy lifting, especially it uses something called VM, which stands for video RAM. The full uncompressed version of Minimax a three requires a massive supercomputer to run, but thankfully the amazing CFUI community has created a smaller optimized version called quantized model. We will understand them when we download them and run it in our computer, don't worry about that. We will talk in upcoming lecture. So even with these small models, you still need a strong graphic card. Minimum to run minimax a locally, you need NVDA graphic card at least 12 to 16 GB of Vim. Example, it would be an RTX 36012 GB or RTX 40 60 TI 16 GB, which I currently have. For a smooth and faster experience, a graphic card with 24 gigabytes of VM is recommended, like an RTX 30 90 or RTX 40 90, rely heavily on NVIDIA's specific technology called QDA to process video. Number two, system RAM. This is the regular memory inside your computer, and for Minimax S three, it is just as important as your graphic card. Because the model files are so large, Cafua uses a clever trick called offloading. Which means it temporarily moves part of the AI models from your graphic card into your computer regular RAM to keep things running. The recommended if you have 64 gigabytes of system RAM is highly recommended. Right now I'm using 32 GB. Your computer will likely run out of memory and Cofi will crash. Number three, storage space. Finally, where are you saving all these file? You cannot run this on an old slow spinning hardware drive. You must use a fast solid estate drive, that is SSD, that particular model in SSD and other files, videos and images files on my regular drive. So I have separated both drives. Because we are downloading multiple AI brains, the video model, the text model and the audio model, you will need at least 45 to 50 gigabytes of free space. But on actual, I should recommend you to use at least 150 to 200 GB or gigabytes of free space. Keep them in your computer. We will not using minimax alone. We will going to use some image models also for image generation. Now, if you're looking at your computer right now and thinking, Oh, no, my PC isn't strong, please do not panic, do not quit the course. In upcoming module, I will show you exactly how to run minimax ST entirely in the cloud using services like Run CFI. You can run a super powerful cloud computer for just a cents and follow along with every single lesson in this course, even if you are on an own laptop, it will work. If your hardware is ready to go, let's move to the next lecture where we will gather all the resources and download our workflow, and I will see you there. 4. Install ComfyUI: First, we are going to set up your computer. I will show you how to install ComfUI and download the Minimax ST files. So everything runs smoothly on your machine. So to install Comp UI, you simply have to type Comfy Dextop on Google, and it will redirect you to your first link that is CompuiTktop. Also, you can download from Github, which is more technical and I recommend you not to use this way. So here are two types to install to come Dektop and install from Github, you have to install once you click here, and it will directly download your EX five. Once you download it, you simply have to run it. Once you go to Python and you have to search on Google and search for Python latest Install. That you have to install it and it will dive to first link download Python. Once you click here, and you have to download the Python Install manager. This is the latest version. You have to install it manager for your Windows. The latest version is install for Windows. That's all you have to do. Now, after installing Python and CPUI, once you download this one, you simply have to download this EXE file, and it will take some time to download the CPUI because after installation, it download approach ten to 15 GB of files from the EXE file. So it will take some time. So once you click here and you have to double click, choose the installation option now, Anyone users with this computer, we have to click this one. Once you click there, you have to press yes, and this is the location you have to install it. Space required, yes, Compi dektop yes. Once I click on Install, it will automatically start installing and it will ask some questions, you have to answer it correctly. That's all. It is about 7.6 GB of five. In your search bar, search for CompuI and it will show you ComfI Desktop, you have to click here and it will open it once you click here. Once you open CompuI you will see this initializing process, and it will take some time depending on the power of your GPU. 5. Download Right files: So once your CUI has been installed and it's open for first time, you will see this window. This is the specific window you will see. Or you will see this type of window that is blank Canvas. You will see this type of error, right? You simply have to create new workflow and delete this one because that was specifically for image generation. Right now we are working on video generation. So you have to go to this template once you click here, and you have to search for specifically if you don't see this window, you see all template and you have to go to Video generation. Once you go to generation type that is video, you will see this type of interface. It can be different for you. If it is different, then you have to search for MINI, that is minimax. You will see these specific three war flow. What you have to do, you simply have to click here. Once you click here, you will see Nothing no error in my workflow. Why? Because I have already downloaded all the files. So I'm explaining you with another workflow, how you can download the error files if there is any if it shows in your scenario. In your window, the error will show in this particular window. You simply have to follow the exact same stat that I'm showing you for another workflow. And you will understand how to install each file and every file for every workflow that is available here. Right now, I'm showing you how to install for video workflow, and then I will show you how to install for image workflow. Steps are same. I'm just explaining you two times so that you can understand it easily. So right now, what you have to do, you simply have to click Image two video, workflow, any of the workflow I'm clicking here. So right now let's use text two Video. Once you click here, you will see this specific error. In the minimax also, this is for LTX 2.5, and we are using here minimax. So we have two workflow that is minimax and LTC. What you have to do, you simply have to click here, once you click here and show the error. Error will show here. What you have to do, you have to download. Either you have to download one by one, or you can download all by clicking here. Once you click here, the download will appear here. That's all you have to do. And I recommend you to download it directly from the CFUI so you don't have to go to technical things like you have to download each file in each specific folder. But I will explain you in upcoming lecture where you can download them and how you can download them just for your information. But the easiest way is that you have to click here and with this one, errors, and with this one, you have to download. That's all you have to do. It will automatically download in the specific folder where it should be. 6. How to download from website alternate method: Within some seconds, it will open. You will not see this type of window right now. You will see this. Once this window will appear, you will see the errors. And once you click here, and you will see all these downloads will appear and you have to download them. You can download them directly, everything at once, or you can download them one by one. It's your choice. Also, you can download them directly from the Internet. Once you copy and paste the link here, it will direct you to hugging phase and you can download all these files. From here, you can download this file, this specific file. And there is one more thing, keep in mind that you have to download them into the specific folder that they ask you to install. Like, right now, I have to download this AEsafner into the specific folder. Once you click here, it will locate you. This is the VAE load. So we have to download this into the specific folder, which it has is state it like we have to download the AI safer, once you click here, you will see the AI Safe ner, we have to download this in this specific folder into the CPI, into the model. In CPI in models into the model into the models and with this one, we have to locate VAE. We have to locate VAE, this is VAE, and we have to download this specific file here in the CI models VAE folder. And where you have to install it files and version. I am going to download this within the CompuI. Once you click on the Download O or download will start automatically, and you can see the download from here. Right now, I'm downloading this one. And once you click T Download, you can see your downloads here. Why recommend you to download it within the Compi Because once you pause, even if you close your computer, restart your computer or switch off your computer, and after some time, you start downloading it again, you can resume it from where you have Blap. 7. Minimax H3 Usage: Let's break down how Minimax a license works in simple terms. First, Minimax release as three as an open weeight model, which is amazing for AI community. However, if you read the official license agreement on hugging phase, like you can see right now here, you will notify specific laws regarding territorial distriction. Due to local AI legislation and compliance rules, four specific regions are excluded from a default standard license that is European Union, United Kingdom, United States of America, and South Korea. So to get approve the application form, don't worry, you can still use by submitting a form, and this is the application form you can apply on. Once you click here, you will redirect to this license request form. You simply have to name it your country and your job title. You have to complete that according to your understanding your company name, if any, otherwise, you can put your name, business email and do commercial products offer revenue. No, you do not have any revenue and confirm, confirm, confirm, confirm, confirm. That's all you have to do and submit that. So once you submit this, you will receive an approval confirmation via email. This shifts the responsibility to you as a creator to use the model ethically while legally clearing you the utilization Minimax S three in your workflow. 8. What each minimax H3 version does ?: Which file actually you are downloading that you have to understand. And if you have a higher version of GPU, then which file you have to download, that you have to understand also. So that the model matches your graphic card. That is the important cache. While I'm using 16 GB of 40 60 TI, that works well with the model that I have downloaded and work well with the model that Cofi itself provided you to download. If you have a higher version of graphic card that it 40, 90 or 60 series, 50 series, higher graphic card, then which models you should download. Let me tell you which you should download. First, you have to understand the file name like Minimax a three, F two val, Pn, Converter, and Sapner. So what is this all about? Minimax a three is third generation model of Hello AI and FL two V is the first and last frame to video and audio. That takes text, one image, and first and last image. Pn is a trim down model that part the model can do without word removed, so it uses much less memory. AND convert is weight squared to eight bit numbers, so the file is smaller. Cort is the specific squeezing method chosen to keep quality and Shaffner is a safe model, ordinary model data Nota program. There are two familiar jobs that we are going to use, that is first and last filame plus audio, that is input text only one image and first last image. And after that matches the first frame and last fim input in our node system. And that different job is the reference to video. That is input up to nine images or three video clips and three audio clips as reference, use it when you want to copy and look character or sound. Next. Same brain is smaller box. So BF 16 is full, 66.3 GB file. Pran BF 16 is compressed, in eight convert is 34 GB of file. PranFPeight scale is 21 GB and Prann in eight convert is R. This is the model that we have used in this particular course. You can test another version of models with your graphic card. If it's work well and good and if it's not, then you can use this that we have used within this course. So what we downloaded, we have downloaded the video plus audio brain, that is Minimax 21 GB, text, reader, video decoder, audio decoder, and turbo addon Laura. Instead of 20 steps, we got 80 steps only. And will it run on my PC? This is the biggest quotien so 66.3 GB might not work practically because of not enough memory. If you have a high end graphic card, high end RAM, then it will work with you. And with 40 GB of file with this one, yes, with Pn, it might work, and with it that is ours, it will work, and it will also work. So if you have a graphic card of six GV and a computer RAM of 16 GB, will it work or not? Nothing will work at all. 30 60 RTX, 30 12 GB, will it work? It will work. With 32 GB of system RAM, you should have 32 GB of SAM with 30 60 tel GB. If you have a 16 GB of graphic card, it will work 30 60 RTX, 12 GB, or 16 GB, it will work minimum with the RAM of 32 GB. This is the model that will work. If you have a 40 90, 24 GB, and 32 GB of RAM, it will work, but you should use this specific model and work with it. If you have a 32 GB of RAM and 50 90, then all these models will work comfortably, and if you have a 80 GB of graphic art, everything will work. Comfortably. If you have a higher end of graphic card with 32 GB, I recommend you at least use 64 GB of graphic card. It's 64 GB of RAM, then it will work really comfortably. Now, if you have a 16 GB of graphic card, that is I have right now and I have. This is the specific case and I have right now, 16 GB and 32 GB, I am using this specific model. Both of these model have worked comfortably with me, but yet it takes a little bit time. If you have higher RAM or higher is specific PC, then it will work competibly even with 96 GB of graphic card and 128 GB of RAM, still, full BF 16 version might work slower PC. Practically, you should avoid this one and you should use and try all these models that is beside 66 GB model. What turbo makes it? Turbo makes it faster. Before we will test it with 20 step, and we will see and compare the time with it. Sometimes it took around 30 minutes up to 30 minutes it took already for me. And with 20 step, it took half the time. And with eight step turbo mode, it took around two to 3 minutes for 3 minutes for three to 5 seconds video. So I recommend you first, try it with 20 step, and after that, eight step or four step, you can try. 9. Choosing the Right MiniMax H3 files: So we are talking about massive studios at with multiple expensive graphics card linked together. Unless you are running a high end Hollywood VFX studio or ranting expensive Cloud server, you should ignore VF 16 files, absolutely. Now, let's move forward with INT eight. To make this model run on normal consumer computer, brilliant engineer quantize the model down into INT eight. That is eight bit ints. They essentially vacuum sealed the math inside the AI print. They also created something called Pune verson. Pruning just means the engineer went in and remove reductant data that the AI wasn't really using anyway and that we call pruned. So the result is pruned int eight files shrink the model down from 66 gigabytes to 21 gigabytes only. Does the quality look worse usually? No. The quality difference between the massive F 16 file and the compressed It file is practically invisible to the human eye, the verdict. So this is the sweet spot. If you have a graphic card with 16 to 24 gigabytes of VM like an RTX 46 TI or 390 or 490, like we have described before, Prone NNT eight model is the exactly fit five for you. You should download this one. Like you can see here, that is a three Prone Neight has been already downloaded here. So this is the suite spot, and here is the suite spot already been installed in my computer. And in your computer, if you have already installed it or run the CVI for the first time, like the previous lecture. So FP eight, this is compression very similar to size to INT eight that is interser eight, but it is specifically turned to run incredibly fast on the absolute newest NVDographic cards. That is, specifically the 40 series and the upcoming 50 series. What I recommend is you use this only. I have tested it. With this one, it is really slow, even for my 40 series card. I don't know why, but it's working. Okay. So if you have 40 90, you know what you are doing, right? You can experiment it FPight for slightly faster generation times. Don't worry. You have already downloaded it. It is just extra information that you should. 10. Using ComfyUI Minimax H3 Video template: So after installing, you will see this specific window. And once you click here, you will see a CompuI. Also, if you see here, you will see update and also click here, you will see CompuICloud. It is tri for free. We will understand it later on. We don't have to worry about this one. You simply have to click here that is Cf UI. Once you click here, you will see Cofi start loading. So Compi is loading. So it has been loaded my last workflow that I have worked on. Let's remove this one and remove this one. Now, once you remove every workflow, you will see this default workflow. Let's understand our CMF UI first. What you have to do, you simply have to go to templates. Once you go to templates, you have to search on video and search for minimax. That's all. And you will see these specific three workflows will appear. First, text to video, second, reference to video, and third image to video. Now, once you click text to video, you will see this workflow. This is specific workflow. What you will see here, you will see the installation guide. Like we have discussed before, you have to download each specific file into the specific folder. Like diffusion model will go into models into the CFUI your install CfUI and within the CFUI you will see this folder model. And within the model, you have to install this specific diffusion model that is Minimax as three in diffusion models. Folder, and here is the diffusion model file. You have to install it. You can see this, the name is exactly the same. After that, the VAE should go into the CPI in models and VAE folder. Again, repeat the staff for the text en Goder in the text encoder folder within the CI models and text encoder, and for the La, you have to install them into Lora folder. If the folder is not visible to you, then you have to create the folder and install them manually. But don't worry. We have another option that is the easiest way. What you have to do, you simply have to go to your error section. Like I have already installed this, right? That is why there is no error or red pop up node appeared. So right now, once you run that once you run this specific template, you will see this type of specific error in that specific template. Right now, it is not visible in me because I have already downloaded this. So what you have to do, you will see this type of error in your work below. Once you click here, you will see parameter node global and errors. Once you click on error, it will ask you to download them. And here is the final download size will appear in this workflow. There is no error because I have already downloaded this. And what you have to do, you simply have to download them. Click here and it will download. After downloading this, what you have to do, you have to refresh it. So all these, these are the nodes that is only for information purpose, and these are the main roads that we are going to work on for this specific model. So what we have discussed in this lecture, we have discussed how to download your specific model. There are two ways to download. First, you can download and keep them specifically manually in specific folder, or you can download them directly from the error section and download them automatically within the folder. That's all. In next lecture, we are going to run this workflow and see what we 11. First video generation: So once the generation complete, you will see in your asset folder, you will see this. You simply have to click this and this is the generated video. It took 151.8 second to generate and you can calculate it. Okay? So basically, it took around 2.5 minute to generate and it's really fast, right? This is the smallest version we have calculated. Only 1 second duration. So it took only 2 minutes, approx 2 minutes. So let's play this. Within 1 second, it generated three shots as per my shot and duration. Now, once you increase the duration and the pixels, let's increase to 0.9 and increase it to wide screen to ultra white portrait or let's make it Portrait white, and the duration should be approach five second. Now let's generate it again and run. Let's see what we will get. Now, our second video is ready, and it is of high quality and how much time it took to generate. Let's see. And it took 2,061 seconds to generate. Let's calculate it, and I will tell you how much time it is. It took about 33 minutes to generate this video, and it is of high quality. Let's play this one. The quality of video is really awesome. Yes, it's really good. Now we have generated our high quality video, right? And it is of five second of video, and five second took 30 minutes. Also, one more thing. It is totally free within our computer, right? So it is totally depend on your machine and which graphic card you are using and which and how much RAM do you have? We will learn to increase the size of our video in upcoming lecture. And also, we are going to learn how we can minimize the time in upcoming lecture. So you don't have to worry about 12. Different generation using other settings: Acute fluffy orange cat sitting on a sunny wood, realistic cinematic shot. Now we are changing duration to 2 seconds only. Let's make it 3 seconds, and also I have increased the mega pixel 0.5 in this particular workflow. Now, after you run this prompt, you will see this is the output that you will get. So now, this is another video. Wow, this is cute, right? Then, again, I have created another video. Ooh. 13. How to change dimention of video: The headin' nach you get arrier feel. It seemed blunt tree, bloodll vomit. Get ready to meet your maker. So all this video has been generated using Minimax, but have you noticed one more thing? We have another dimension of video, and this is another dimension of video, and this is another dimension of video. This is 169 video, and this is a real format or short format video within mobile format. And now this is square format. You will see. Now, let's play this one. All this video has been generated within Minimax. So all the video that you are watching right now, it's created within minimax. These are really amazing. So to change the dimension of your video, we need to change a simple settings. That is, we have used a mega pixel. Right now, you can make it to nine or you can make it to four. Now for this one, you simply have to change. According to expectatio, you have to change the dimension. Like, hey, you can see expectatio. Let's hide this one now click here. Once you click here, you will see a square video, a portrait video, photo video, oh it standard or ultra wide or portrait white screen. All these formats are already available within Minimax. Once you change the dirention and hit run again, it will generate a video, but not the wide one, but in a real or short form format. I have tried to make it on a large scale, but it's not working at all for me. Why? I got the memory error every time. So more than nine, I got the error. You can make it 1.1, 1.2, 1.3, 1.4. It's totally depend on you. For the megapixel of the video. So the best strategy to generate you test is a lower resolution, like seven, like we have already generated within 0.4 megapixels. That is 864 to 780 megapixels. Later on, we will learn how to use AI upscaling to make them super sharp or focal equality. So don't worry about the video if you are creating it in smaller version. Now that you know how to shape your video and avoid dimension error, let's move to another lecture. We are going to look how to control the length of your video using 24 frame per second grade. See you there. 14. Control duration of video: D. Welcome back. In last lecture, we have learned how to change the shape and resolution of our video. Now let's talk about time. How do we control exactly how long our video clips last, right? In this lecture, we are going to look at the duration settings and 24 frame per second grade. Let's understand the basics. How long can we go? Minimax, a see is designed to generate video clips up to 15 seconds long. Wow, that is really long. Now, but not long enough like seed dance 2.5. Recently, we have learned a seed dance 2.5 can generate up to 30 seconds video. That is really Cool. Now, it also generate this video at a smooth cinematic frame rate. It's 24 frames per second. That is 24 FPs. This means every single second of video, the AI is drawing 24 individual pictures. If you want five second of video, the AI generated 120 individual frames, if you want to 15 second, if you want to 52nd of video, it has also generated 360 frames. Wow. So right now, A has generated about 72 frames of pictures to generate this particular video. Let's run with another prompt. Meanwhile, let's understand what we are looking at. In this standards, basic Cf setup, you usually have manually type the exact number of frame you want. But with minimax ST that gets tricky. The AI actually process frames in every specific mathematical block. Technically, a 17 frame per block grade. If you type a random number of frame that doesn't fit that exact math, the generation will fail and give you an error. So I recommend you don't change any value here. The custom template we are using a a mathematical expressions, mode hidden under the hood. All you have to do is click the button of three to 5 seconds or 10 seconds or 15 seconds and choose a template or use a prompt. That's all you have to do. So we have to understand time versus rando speed. So why not just set every video to 15 seconds? The answer is rando time, right? Unlikely resolution where making the videos larger makes the generation time spike unpredictably. Duration is scales in very state linear line. If a three second video takes four to 5 minutes in my specific machine, it might take different time for you. So we are living in different time zone in generating videos right now. So for six second video, it might take up to 20 to 25 minutes for me. For 15 seconds video, it might take approx one for me. So I usually create videos under three to 5 seconds that work really fast. On the ten to 15 minutes, I always get the results. But sometimes I get it really fast. So the basic strategy for beginners is that 15 seconds can take a long time to render. Hey some professional workflow advice. When you're testing a brand new idea, testing a new correct or trying out a new prompt, always set your duration, three to 5 seconds. That's all. This allowed to get a quick preview. You can see if the lightning looks right, if the corrector matches your vision and audio sounds food. Once you are 100% happy with that five second test, then you can change the duration to ten to 15 second depending on your users. Click generate and walk away to grab a coffee while your computer enters a final masterpiece. That is totally free, right? Now let's zrap up. We now know how to set up our workflow, control our video shape, and set our duration. So in the next lecture, we are going to learn how to actually talk to AI. We will dive into prompt engineering, and I will show you how to write instruction to perfectly that the AI gives you exactly what you want every single time. I will see you in the next lecture. Thank you for watching. 15. Understand the prompt : So right now, there is a prompt. Let's understand the prompt one by one. Okay? So let's first watch this video. It is amazing and satisfying video. Now let's understand the prompt. So up until now, we have been focusing on the button, the settings, and the meth. But now we are going to learn the actual language of Minimax as three. That is we call prompt. Let's understand this prompt. It has established the rending style immediately to pixar three D animation rather than letting the AI default to photorealism. Here, the key details terms like subsurface scratching, squares and stress and dreaming bouquet. Tell the model how light depend and character movement should behave. It gives the core narrative an emotional context before breaking down into individual moments, establishing the relation between moves. What is the scene overview here? It gives the AI the core narrative. And the emotional context before breaking down individual movement, and now the storyboard. It leverages Minimax ST native timestamp features, execute three distinct camera cuts within a single three second generation. Close up on the girl, roll the ball, action shot of the kitten, bouncing with squares and sass motion. Wild close shot of the girl hugging the kitten and the exact same duration of the storyboard. That is on zero to 1 second, it will happen 2 seconds this happened and 3 seconds in this happen. In 3 seconds, white close shot of the girl hugging the kitten, that this happened. What about camera and negative routes? We have instruct the AI on how to handle transition and previous unwanted artifacts. Ask for clean, soft rapid cuts rather than cross fades and explicitly ban text logos and photorealistic texture to keep the stylised three intact. And at last, we have audio layer. Take full advantage of as three native audio generation, combine diactic soundscape elements. Example, giggles, meow, chirping birds, rustling bees with a known disactc core and orchestral melody at the background, and a magical chime on the hug. Now, instead of giving vague exactive, we have given Minimax X three a complete director sheet, artist style character motion, exact frame by frame timing, camera angles, negative constrat and a complete sound design, all in one structured. 16. Generate cut scene: And of world. Go. So in our last video, we have learned a four pillars of strong prom. But we only shot a single continuous camera angle, but if you want to show a wide shot of a car driving, then instantly cut to the close up of the driver phase within the exact same video. Traditionally, you would have to generate two separate videos and state them together in editing software. So Minimax a three is so advanced that it understand multi shot sequence right inside a single tax prom. We do this using something called explicit timestamps. So we have to understand the timestamp first. It is a simple tax bracket telling that the AI exactly when a specific action or a camera cut where it should happen. Instead of writing one long paragraph, break your prom down into individual shot and place the time bracket at the very beginning of the sentence. For example, if you're getting a six second video, you might want the first three second to a white shot and the last three second to be a close up. You would write it exactly like this. 3 seconds a white cinematic sort of a red sports car driving down a desert highway, the camera track alongside the car and second sort, a tight close up on the driver face where he's wearing sunglasses, smiling. The camera is mounted on a dashboard completely static. When you use these S brackets, the EI understand that exactly three second mark, it needs to perform a hard seamless camera cut. It won't bend or mob the images together awkward. It will cut cleanly just like a real movie. The biggest advantage of doing this inside Minimax Steres is how it handles the audio because the AIA generating the native stereo audio simultaneously. It keeps the background sound continuous, even when the camera angle changes. If you add your audio cues at the bottom of the prompt, the sound of the car engine will roll smoothly across the entire 6 seconds completely unaffected. By the visual cuts. When you add as many cuts as you want within 15 second limit, you could do three, five second shots or even rapid three second shots for a high energy action sequence. Just remember one golden rude. Always keep the total time in your prom, match perfectly the duration node settings in CI. If your prom describe ten second sequence, but your Cui duration node is only 5 seconds, the AI will comprise the action of your video within that 5 seconds, and it will look chaotic. This out yourself in comfy workspace, set your duration to 6 seconds and write a simple two short prompt timestamp bracket. Once you have practiced that, you might realize that writing all these brackets and camera direction by hand taking a lot of time. We will try to automate this entire process using a custom AI assistant, save in the next lecture. 17. How to create prompt using Ai: So by now, you have learned how powerful minimax S three prompt can be, especially when you include style anchors, subject mechanics, camera moves, timestamp, and dual layer audio. But let's be honest, write out of 150 word highly structured prompt by hand, every single time you want to generate a five second video clips is tedious and time consuming. If you miss a bracket, forget the audio text or describe some too vaguely, the AI might not give you what you want. So in this lecture, we are going to learn how to automate the entire prompt creation process using AI. We are going to build a dedicated custom GPT or system prompt for Cloud, hat GPT, Gemini, or any other chat bood available on the Internet right that turns your simple lazy one line ideas into perfectly formatted cinematic minimax as prompt in seconds. So why build a dedicated assistant? Think LLM like JAG Pitt as a junior screenwriter. If you just ask had Gi Pitt, write me a prom for a video of a robot. I will give you a generic overly wordy paragraph full of useless buzzwords like hyperrealistic, ultra and photorealistic. As we learned earlier, Minimax ST doesn't care about those buzzwords. It needs camera angle, lighting, direction, physical movement, speed, timestamp, and audio cues. By creating a custom GPT with a custom system instruction, we lock the AI into a strict formula. You give it a rough idea like a cyberpunk Samurai standing in the rain, and it instantly outputs a production ready script format is specifically for Minimax as three. So let's build this together step by step. If you have a Cha GPT plus account, follow along in the Explorer GPT free version of CHA GPT GeminioClod. Don't worry, you can simply paste the system prompt. We create directly into your normal chat window. That's all. So let's open CHARGPTPaste this. What is this name, Minimax SE prom director description, convert rough videos ID into production ready Minimax SD prom with visual mechanics, camera move, timestamp, and audio cues. So and third step in this instruction text box page the following master prompt template. This is the exact brain formula that teases the LLM how to format prompts for Minimax as three. Let's copy and paste. Simply run it. So now let's test our new assistant with R world example, so you can see magic in action. Let's paste this prompt, that is, create this action luxury commercial sort of espresso pouring into a glass cup in N Marvel counter. That send it. And what har JP will give us. So Chair GPT has given me an image. It's awesome. No worry, but I don't want the image, right? So you have to specify that you want the prompt here. Let's sand prompt again. It had given me something that I can copy and paste it directly. In the comfy, let's paste this one, and in the comfy, let's pair this one, and this is the audio subject mechanic anchor settings. Now, let's change the duration to 3 seconds and the megapixels to something in the middle. Let's run it again. I can't imagine that Miniax had created it and it's really awesome. It's look like someone has actually recorded this video. The details of pouring this coffee into the cops really amazing. And it's to the point, gravity is working perfectly fine and the camera that it has explained is really good. One more thing, it's actually copied this angle. It looks amazing. But at the video point of view or advertising point of view, this angle is also good. Where it will help you. Imagine you have a quick client request for a luxury coffee commercial, but you don't feel like typing out all the camera and audio details. 18. Generate Pixar Style short animation: So this time, we are creating this type of animation. For this animation, we are using Chat GPT again. In hat GPD, I have asked. This is the same chat that I have used before with my prompt, right, prompt guide. So it is just a free version. You don't have to pay for it. Right now, what I have used, I need a five second Pixar style, a cute robot in a dusty, attic pints and all glowing light bulb, screw it in its chats and I light up happily. So give me prompt. This is all I have asked. So hat JBT has understand about our previous chat and gave me a prompt. And this time, I have simply copy and page the prompt here. Change the duration to 5 seconds. After that, for the resolution, I keep it same. Now let's render this one and see what we got. Now, after rendering it, this is the result that we got. It's really amazing. By delegating the writing process to a custom like ha GPT LM assistant, ensure 100% syntactic accuracy for Minimax a three and speed of your production workflow tenfold. That's complete our lecture on prompt engineering and LLM assistant. You now have total mastery over text bikes, video generation, and prompt construction. 19. Using Skills Create Product Ads Video: So we are getting into the advanced version of Minimax as three, where we are going to use something called skills. These are the specific set of prom that works really well with minimax. First, what you have to do? You have to search Minimax Aster skills on Google, even if you don't find, A, you will get the link in the resource section, don't worry about that. Minimax Aster skills, once click it and you will scroll down and you will see so many examples that has been created using Minimax as three. So what we are going to do, we are going to use this specific skill set right now. There are two types of skill set. First, the text base and second the reference space. First, we are going to use the text baase. Once you click it, you will see a code or blame. All these code that you can just copy, just like select two lines that copy or you can download them and use it as a scale in Cloud. Don't worry. We are going to use free version in this course, so we are going to simply copy this and paste it in the geminiOce you paste it, and in the last lines, you simply have to write this simple prom that you use this skill and give me prompt for Speaker's product ad. That's all you have to do. It will not ask any reference for you. Instead, emina want me to adapt this into 12 VA prompt, which is reference to image prompt with specific product reference. Right now, we do not have any specific product reference, so we are going to create our imaginary product, that is speakers and use this prompt within the CI. Let's open the CUI. Once you click here, you will have to search CUI, and that's all. CFUI dextro will appear. Even if you don't find it, you will find it in the Dextop. Click on Come PUI, maximize it and wait for the installation. So once you open it, you will see Minimax, and this is your template. Even if you don't see this, don't worry. Once you click here, you will see this. Don't worry. You simply have to go to templates and click on this text two video template. And here's the prom that you have to copy. Paste it here. But here's the catch. You have to understand the bracket system that we have already discussed the sort one, live action, microfilm, and you have to understand the sort two at 3:55. That is, we want it at 250, zero, the camera cuts thub T S one enters and it happened at shot three, that is, make it pp. So we are going to get this five second video, shortcut, and subtle rhythmic overalls done. So this is going to be a five second video. We are going to create the megapixel to nine. That is, it will be uh, 720 pixels video, and we are going to use turbo mode, that is, make it true, so it will be fast and step we are going to use here is eight. Now with turbostap eight, now let's run it. Once you run it, you will see magic. So the video has been generated, and let's play this one. But let me tell you where this video has gone. You can search in the assets, and this is the video. You can download it. Again, it took about this is 720 P, and you can see the quality is better, how much time it takes. So it took 15 minutes for this video to generate where you have downloaded the UI, and in the Comp UI folder, you have to check for the output. In the output, you have to check for the videos. Here is your video. Now let's play this one, and it is about 1.15 MB. Pure sound, untethered. It's great. Just like a complete sequence of video, but incomplete also, right? This is on you. Copy this prompt, paste it in your prompt section, and show me the output where you want to show me. It's up to you on social media, anywhere you can share it with me. I will comment on it, and we are going to move forward with another prompt. 20. Using Skills Create Product Ads Video 1: So we are getting into the advanced version of Minimax as three, where we are going to use something called skills. These are the specific set of prom that works really well with minimax. First, what you have to do? You have to search Minimax Aster skills on Google, even if you don't find, A, you will get the link in the resource section, don't worry about that. Minimax Aster skills, once click it and you will scroll down and you will see so many examples that has been created using Minimax as three. So what we are going to do, we are going to use this specific skill set right now. There are two types of skill set. First, the text base and second the reference space. First, we are going to use the text baase. Once you click it, you will see a code or blame. All these code that you can just copy, just like select two lines that copy or you can download them and use it as a scale in Cloud. Don't worry. We are going to use free version in this course, so we are going to simply copy this and paste it in the geminiOce you paste it, and in the last lines, you simply have to write this simple prom that you use this skill and give me prompt for Speaker's product ad. That's all you have to do. It will not ask any reference for you. Instead, emina want me to adapt this into 12 VA prompt, which is reference to image prompt with specific product reference. Right now, we do not have any specific product reference, so we are going to create our imaginary product, that is speakers and use this prompt within the CI. Let's open the CUI. Once you click here, you will have to search CUI, and that's all. CFUI dextro will appear. Even if you don't find it, you will find it in the Dextop. Click on Come PUI, maximize it and wait for the installation. So once you open it, you will see Minimax, and this is your template. Even if you don't see this, don't worry. Once you click here, you will see this. Don't worry. You simply have to go to templates and click on this text two video template. And here's the prom that you have to copy. Paste it here. But here's the catch. You have to understand the bracket system that we have already discussed the sort one, live action, microfilm, and you have to understand the sort two at 3:55. That is, we want it at 250, zero, the camera cuts thub T S one enters and it happened at shot three, that is, make it pp. So we are going to get this five second video, shortcut, and subtle rhythmic overalls done. So this is going to be a five second video. We are going to create the megapixel to nine. That is, it will be uh, 720 pixels video, and we are going to use turbo mode, that is, make it true, so it will be fast and step we are going to use here is eight. Now with turbostap eight, now let's run it. Once you run it, you will see magic. So the video has been generated, and let's play this one. But let me tell you where this video has gone. You can search in the assets, and this is the video. You can download it. Again, it took about this is 720 P, and you can see the quality is better, how much time it takes. So it took 15 minutes for this video to generate where you have downloaded the UI, and in the Comp UI folder, you have to check for the output. In the output, you have to check for the videos. Here is your video. Now let's play this one, and it is about 1.15 MB. Pure sound, untethered. It's great. Just like a complete sequence of video, but incomplete also, right? This is on you. Copy this prompt, paste it in your prompt section, and show me the output where you want to show me. It's up to you on social media, anywhere you can share it with me. I will comment on it, and we are going to move forward with another prompt. 21. Short Animated video: So we have used this type of animation. Now it's time to use this type of three D animation. Simply click this one, and so you have to click this one, copy this, go to code, and copy this one. Now let's copy. Chat GPT, let's ask. Let's pace the skills that we have used, use the skills and give me complete story under 5 seconds with five with cut in every second. Give me a new story, prompt for it. Now let's try it. With this one, let's copy and paste this prompt, five second cut and copy this prompt and paste it into the a specific area in CFI. Now let's pace this one. And with this one, let's try to make it a smaller one and run it for 5 seconds with moto and run it. Now the video has been generated, let's watch this one. So the video is really amazing and it actually tells the story that star come and it goes up. So that is we can say Twinkle, to tinkle little star. So it's just a simple story, yet we have got a really, really amazing video. You can see cut at 1 second. 2 seconds. The second cut. 4 seconds. And yes, it's complete. So we have four cuts in this video, and it's working really fine. Let's move to our next video that we are going to create. 22. Paper craft stop motion Animation: So next move to another part of the video. We are going to create this type of animation. Now let's click this scale. Again, the same step code, and copy, paste this here. Now, let's paste it here. Different style of animation. No, not style we are going to use. Create another story skills four under five. So let's create another story with this one. Give me prompt. So before we have created Pixar, inspired three D animated short film, and now we are going to create another story, the paper, but first, now let's copy this prompt and paste it into our Cf UI. Again, fui has been done and now paste it here. Run again. So another video has been created. Now let's play this one. It is really amazing video. The clarity, I can tell that this is, as far as I can understand, Minimax is the best in video generation right now. I have tested so many things, but Minimax has actually the upper hand. So this is the video that has been created, and it is actually amazing. 23. Generate complete Brand promotional video: I Now we are going to create another video that is brand promo video generator. So now let's again, click on Skills. Let's close another tab that we have opened before and go to code again and copy and paste it here. So this time, we are going to create a longer video. Why? Because we want it to be amazing video, right? So let's open this one and see with this one target duration normally 15 seconds. Recommended 15 seconds when users want a fast launch film. So they ask us to create a 15 second video, but let's try it with the 10 seconds video. In the HAR GPT, again, paste this one, create four. No. I think something has let's select and copy and paste it. Now with this, create ten second video, for a promotional video. Cuts depend on you where you can give. And also, I want it to be very interactive retention friendly. Now let's make it. With this one, I will need a prompt here. 10 seconds video. Chat GP has asked me, where do you want to create a video, right? So let's make it this one for homemade Kitchen. So I'm creating this video for homemade Kitchen and video format, let's make it short or YouTube reel video format or YouTube video. Now, yes, and make campaign focus, product or service launch, and showcase feature and benefits. No, don't worry, we are going to product and service launch, right? Now, let's done. With this one, we are going to get a prompt here. And these type of images, Wow. It is actually going to create a video for me copy complete video prompt. Now we are going to copy this prompt. And they have created in Hindi, according to the video genre, I guess. So let's paste this one and make it friendly woman paste it here. And now with this one, I pasted everything, and let's make it in English with order today. That's all. Now let's make it to 10 seconds and now run it again. With this one, let me see if it is yes, format, Instagram real. So this is going to be a Instagram real, I guess, yes, yes. So let's change the aspect ratio to 916 so it will match our video perfectly so that it will not conflict with our prompt and the video. Now let's run it again. So the video has been created. So the video has been created, but let me tell you, I have tried it first with the ten second and with the 0.9 value, but it didn't work. I don't know why, but my system or my graphic card is not actually working at all. So then I tried to reduce the Aspecraso to 5.5, then it's work. So now let's see how much time it took. It took around 10, three 7 seconds. Wow. Now let's play this video. Missing GurkakanaFreshly made homemade with love. Pure foro tanchaPure comfort. Gur J Sasa kitchen. This is what I want it. This video is really great. Yes. But we want the same output like the one we already have include it. We are not going to slack this one, right? So let's brand promo video we are going to use. Let's copy this one, paste it here for marketing agency. Now let's render it one once again. So it's not working at all, right. So now let's ask Claude here information that we have past it now. I want using this skills, I want promo video for agency. We can say marketing agency or we can say video, AI video editing, AI video. Generation agency. Now let's run it. We have done a little bit of mistake. Let's make it. So I want for ten second video. That is 416 ratio nine and second quien promo quality, range of uses, add film content before or after replacing expensive tool. And we can actually tell if it's brand promo for market and creative production promotional brand promo video, yes, this is the brand promo video. Visual style, I want this type of visual style, right? Which type of visual style we want it. We want this one. Promotional content brand website, shop, personal skills organized based on Provin, motion graphics, and I want it for create in general agency, not specific color. Four third question, you decide and let's see what Claude will decide and give me prompt. Now let's copy this one and paste it here. Paste it. Now again, render this one with a ratio of 16 and now run it again. 24. 25 Google style promotional video motion graphics: In this particular strap, I'm using the same. Everything is same. I have only asked Chair GPD this time that minimax X free video generation Google style course promo. This is the video that I wanted it to generate this time, and it has generated the code for me. You can say a code or you can say prompt and simply copy and paste it. Same step, nothing has changed. Duration you can see is 0.5. Now, duration, no, mega pixel is 0.5. That is, again, in this ratio, duration this. Now this is the animated video and do not use Google Logo branding. Use Okay. And after that, it has generated this ten to 12 scene. Yes, perfectly aligned. No, now let's run it. Wow. It has actually created the video that I never imagined. And the animation is so good that I want to capture it in a loop. So don't worry. It is the retention video, right? It is kind of a retention type video that always Google make. But it is not just like a TD style, as you can see, this is not just like a three D style animation, right? It is like, Google actually make these types of videos. But they Google now use simple illustration, not TD style animation, but, yes, the video it generates, it's up to the mark. But the problem is with this that the text, this is the problem. Animation is so good, no issue with that. If the text size is bigger than the animation, and the video is great, but if you use a text with a smaller size than it is not able to create it. So we can actually not using text, right? We should avoid generating text. 25. Method 1 Video upscale: We are going to talk about the first method of installing upscaling the video. How you can do that? First of all, in CVI, we have to go to extension. This is the important part for this method. You have to install the extension. That is NVDA. First of all, extension extension, you have to search for RTX above. Once you click here, you will install it and enable it. Make sure it's unable. After that, you have to refresh it and we are going to create a new workflow how we can do that. You simply double click on the Canva and search for the load video. Here is the load video and you have to use the CUI essential od video. You have to search for another node that is Get video component. Once you click here, you have to connect it all the nodes with video to video. After that, you have to node that is RTX video super resolution and connect the images to image node. It will automatically connect and after that, double click on the Canva, search for the create video, once you click here, you will see and again, connect. Or the nodes that is upscale image to images. After that, audio to audio, just like you can see how I've connected, and after that, you have to check for the save video. Once you connect it, video to video has been created. Another setting that you have to check that is encoding to re encoding and CRF to value of six. Once you select these settings, you have to select the video. After that, you have to run the Park flow. That's you have to and you will get the result. But I believe that I don't see much of a change in the video section. On the left hand side, it is the actual video that we have created, and on the right hand side, it is the upscale version. So I don't see a much difference in the generation with the upscale. And later on, we are going to see the actual upscaling of the video that we can actually sport the difference, the quality and the sharpness we will get in the upcoming upscale lecture. 26. Method 2 Video upscale : So this time we are talking about second method in which first you have to go to this time in the templates, and in the templates, on the window pop up, you have to go to utility. Once you go to utility, you have to. You will see there are several nodes system. So what you have to check, you have to check for the specific upscale video that is C VR 23b. Once you click here, again, template will load, and what you have to do, you simply have to again, the same step you have to follow. That is, you have to click the view details in the error found, and it will show you some file that you have to download, and these are the missing inputs. Once you click on the Download, all the models that you need to download, it will download it. Automatically in this specific folder. So it is about 3.69 GBOFle. You have to click Download A, and they will automatically download for you. See the download has been started, you have to go to view A and there are the download. Once they are downloaded, what you have to do, you have to check for the after the download, you have to check where they have downloaded. You can check them. So in Explore and here it is. This is the specific folder that they have downloaded the Cofi in the diffusion model. After that, you have to click on Refresh, or you can simply restart your ComfuI if they have not updated in your note system. The error has been gone, and now they want us to load an image or video that we have to upscale. Not specifically for image, but for the video, right now, I have selected this video. The settings here. You can simply change the scale multiplier to 1.5 or 1.25 or you can keep it as it is. Right now, I'm using scale multiplier is two, so it will double the size of my video in the pixels. Also, it will take so much time according to the generation or the graphic card that you have. For me, the generation has failed because of the not enough GPU memory, as I've told you, at first time, I've tried it, but after that, I have to check for these specific settings. You simply have to follow what I'm looking, what I'm changing here. That is trim video. If you need to trim the video, you can do so. With a split ten it was false, then you have to click it and make it true. Run again. Scale multiplier to 1.5, it's totally depend on your GPU power. So the generation is complete after running it. Now, let's watch the difference. 27. Image to video generation: Once you open your confi again, you will see the same video that you have used last time, right? Let's move to templates. Once you click on template, you will see so many templates are available. What you have to do, you simply have to go to all templates. Or you can go to simply Video template. Once you click here, and you have to search for minimax. Once you type Mini, you will see three specific system here. So first, text two video, we have already used it. Now we are going to use Image two Video model. This is the specific model that we are going to use. Once you click here, you will see input media missing. This is the error you will get. So here is the media that we want to upload and it will automatically generate with the image, video, prompt and save video. This no system it's exactly the same only they have used first fim here. You can see here is no first or last film as an input here, but here we have first and last film as input here. Click here or you can drag any of the image. Example, select any of the image like this one, I'm using this image. Now our image is open. Expectato I'm using square as per our image that we have selected. And with this one, I've done nothing. I simply change the duration to 2 seconds only and with mega pixel 0.4, and now, what I'm doing, I'm doing simply run. I have done anything change. I have not done anything change here. Run it. But before you run, you might see this type of error to resolve them before running the workflow, you need to remove all these error. How you can do that, you simply have to click on View Details. Once you click on View Details, you will see the error. This is the same step that we are going to use that we have to use the same step that we have used for this one. Taking an example of LTX 2.5 image two video model. Once I click view details, so you have to download it. After downloading all the files, your workflow will work. Same for each template. This time we are using Minimax, you have to use once you click here, you simply have to, and let's run this work flow. Run it. 28. Create Video Sequence using Single Image: So what you will get, let's run this one. So we have got this amazing video, right? It is actually following the prompt really well and not following the reference image. Why? Because it is not actually related to this. So we are going to create another frame that is what I did this time, I have simply copied this prompt and pasted it into Gemini. Once I paste it in Gemini, and in the end, I simply ask Gemini, this is sample prompt. I want prom for uploaded image, and this is the image that I have uploaded. This is the image. Now, once you scroll down, you will get this prompt. You simply have to copy and paste the prompt here. Once you paste it and run, again, for three to 4 seconds, I'm running it around. And now let's simply run it. So now let's check out the video that we got here. So this is the video that we got, and it's amazing, right? I never thought that we can create a video like this. Amazing. 29. Ad Product Video sequence: This time, I'm using this product, and I have uploaded this in the Google Gemini again. And now let's make a ten second video, promotional video or product video following the script style I have shared before. Nothing has changed. Now let's Enter it. Actually, Google Gemini is starting to creating the video 10 seconds so that I don't want it. I want this to be a prompt. Now let's run it, and let's see if it will give me prompt or not. It just get hallucinate. Right now, let's This is the scene three second video. No. I want more shots. I want more shot video. Make it at least. Five shots in 10 seconds. And I hope we got something that might work. So let's copy paste the exact prompt here, and also let's change it. Simply drag this image here. And right now I want this in nine, and I want it to be with at least 10 seconds, and let's see if it will work in turbo mode or not. Let's run it. So this time, we got the error of not GPU memory. So this is what you get when you use higher memory duration. So let's change the duration or let's change it to maybe 0.5. And now let's try to run it again. And I hope this time it might work. 30. Avoid mistake before generation save time : Now video has been created and let's see how much time it took. It took about Wow. Let me tell you one thing. For just two or 3 seconds video, it took about 700 seconds. Right now for ten second video, it took under 600 seconds and how much time will it take? It took about maybe under 10 minutes. Under 10 minutes it's done. Wow. Now let's play this video. So video has been created, but the problem it is actually not using my product that I think it should be. Why it happened because of the gemini. Let's first read, and this is the problem that always we get, and this is what I forgot this time to explain that we should at least read our prompt first, even if it's using or not, the same product or not, right? So this time, Wireless gaming headsets, it's talking about. This is really bad prompt, and this is not what I was expecting. Video that we got here is really amazing. It's awesome. I can't deny this one. But problem is with our prom. Use this prompt, like, give me prom for this product that is speaker I'm uploading the image. And now let's it should be under five second video and five shots. Submit, copy, and paste it here, for the frame, yes, and for the duration, let's make it to 5 seconds. And with this one, editorial text wireless speaker the side of soft texting by thematic, yes, yes, yes. Run it. And let's see what we got here. I hope this time we will not get something different. Now let's run it. Okay. Okay. 31. Get different result each run time: Heads of clapping that we got the result right. Now let's understand one more thing why we not get the result before. Because also we have not used this tag also, even in this prompt. And somehow MiniMax, it's really powerful to understand that and give us result according to our speakers. And right now, let's see. This is the main thing that we wanted to add, according to the first image that according to the first prompt that we have added here is C. This is the picture of the mouse gaming mouse picture. The first prompt whenever you try to load the template here from the template sexton, you will see example, you will see once you go to template and load the template, you will see a prompt here with the image here, once you click it, you will see this with this one, you can see here is the itorial picture mouse from picture one in its original scene. Picture of the transparent mouse, it will automatically understand that, but we have to refer it in our prompt so that MiniMax can understand for which product or for which image we are referring to. And this time we are referring to this image. Seconds run it with a different noise. I'm running it again. Now let's run it. We got the error of not GPU memory. So this is what you when you use higher memory duration. So let's change the duration or let's change it to maybe 0.5, run it again, and I hope this time it might work. 32. Project: Oh, that's it for this course. I know you have learned so much. I want you to create a project on any of the video that we have created so far. You can use the skills that we have learned or you can create your own prompt and try it in the Minimax within Comfy hit run and create your own video and share it in the project gallery. I would love to give a feedback on it, and I can't wait to see what you have created. See you next time. Thank you for watching.