AI Character Consistency: Same Face, Same Outfit, Every Image

1. On_Model_Course_Trailer: Hi. I'm Sakaia. For a while, life kept me away from making courses. That's behind me now. From today, I will be right here with you, learning lots of new things together and this course is the first of them. In this course, we are going to learn how to keep an a character looking like the same person across 50 images. You will write a one page Character Bible, build a frozen set, paste a log block, and check every image with a seven leaf test. By the end, you will have six poses of one person, not six cousins. Let's begin. One more thing, one more thing, I used Lenaps to make my voice clearer and Google Flow for the small animated clips you will see along the way. What you'll be able to do. By the time we finish, you'll have written a one page Character Bible for a character of your own. Face, hair, signature item, wardrobe, props, palette and poses all pinned down in plain nouns a machine can't wriggle out of. You'll also have described a frozen set, a room built once and locked. So the plant, the mug, and the light stay exactly where you put them. You'll be able to write a locked block. The fixed prompt structure I paste into every single generation and change just one thing in it at a time. You'll run the seven leaf test on any image and name the kind of drift you're looking at, whether it's the face, the clothes, the room, or the light, and you'll know the fix for each one. Most of all, you'll walk out with six on model poses of the same character side by side that look like they came from the same hot. Six becomes 54 with a bit of arithmetic, and I'll show you that too. 2. Opening: I'm Sahaya Babu. I work in IT, I design and illustrate, and I've spent years tutoring people through exactly this kind of thing. These days, a lot of my week goes into explainer videos for a customer support software product. Every one of those videos stars a character I have to keep the same from the first shot to the last. So let me say your problem out loud because I've had it, too. You make a character you love. She's perfect. Then you ask for her at a desk, and a completely different woman sits down with a new face and new hair. The cream jumper is now a gray hoodie. You try again and again. And by the tenth image, you've got ten cousins and no character. That's the bit that makes people quit. Here's the thing. It isn't luck, and it isn't that you need a better tool. It's that nobody's told you the model forgets everything between images. Today, we fix that with a method, not with crossed fingers. I'm going to teach the whole thing through one real character of mine, a support agent called Anika. She has black curls in a messy bun and a cream turtle neck, and she works in a room I call the Golden Hour home office. You get to know her desk better than your own. If you'd rather use your own character, your own style, or a different image tool, please do. Anika changes, but the method doesn't. By the end, you have a one page Character Bible and six poses of the same person, not six cousins. One person six times. 3. What_You'll_Need: We'll spend about 6 minutes getting the desk ready, and it's a short list. First, an image generator you can type into. I move between Chat GPTsimageGenerator, ideogram, and Google flow depending on the job. Mid journey works. Whatever's already open on your laptop works. If your tool lets you attach a reference image, brilliant. We'll use that later. If it doesn't, you'll still get there. The words do most of the lifting today. Second, somewhere to write that isn't the prompt box, a plain document, a Notes app, or a sheet of paper and a pen. I use a plain text file called bible dot TXT because plain text pastes cleanly anywhere and never sneaks in odd formatting. Think of this as the place the character lives. The prompt box is just where you visit her. Third, a folder, one folder named after your character with nothing else in it. Mine says Anika, and that's it. Every image you make today goes in there, including the bad ones, especially the bad ones, actually, because we'll be comparing. Fourth, a character. If you already have one, bring her or him along. If you don't, pick something simple for today, a baker, a delivery rider, a school teacher, a cat in a raincoat. Keep the outfit to three or four pieces. A wizard in seven layers of embroidered robes is a lovely idea for next month. Today, the fewer pieces, the fewer things that can wander off. That's it. No paid upgrade, no plugin, no secret settings. Set your tool to a wide frame if it asks. 16 by nine is what I use for video, and open your blank text file next to it. Left half of the screen for writing, right half for generating. That split is going to matter more than you'd think. One small warning about the tool because it trips people up. Many image tools quietly rewrite or expand what you type before they draw it. That's fine for a one off picture, but it's one more reason to keep your own master copy outside the tool. Think of it like a recipe card taped inside a cupboard door. The cook can improvise all they like, but the card stays the same. 4. Project One: Your_First_Go: Right. Before I teach you anything, I want you to make something. We've got about 10 minutes for this, and I'm going to be annoyingly unhelpful on purpose. Here's the brief. Generate your character three times, once as a head and shoulders portrait, once sitting at a desk and once walking. Write whatever prompt you'd normally write, whether that's short, long or downright poetic. Just do what you'd do if I weren't standing here watching. I'm not going to tell you how yet. Have a go, get it wrong, and then the next hour will actually mean something. If I give you the method first, it's just a list of rules from a stranger. If you feel the problem first, every rule lands as an answer to something you just watched happen. And it will come out inconsistent. I promise you that. When I first did this with Anika, the portrait was lovely. The walking shot gave me a taller woman with straight hair and a blazer I'd never asked for. Mine was terrible, and so is everyone's. That's not failure. It's just the starting line. You've got your three, drop them in the folder and put them side by side. Now look hard and ask one question. What changed that bothers me most? Maybe it's the face, maybe the outfit swapped. Maybe her hair went from a bun to loose. Maybe the room is different every time. Pick the one thing that annoys you most and write it at the top of your text file in your own words. Keep those three images and don't delete them however ugly they look. At the end of class, we're putting them next to your finish set, and that comparison is the best moment in the whole hundred minutes. Off you go, you've got 10 minutes and I'll wait. Done? Good. Read me what you wrote at the top of your file. I bet most of you wrote something about the face or the clothes. A few of you wrote the room. Hold on to that sentence because the next three parts are going to answer it one by one. 5. Part 1: Why_Characters_Drift_Character_Bible: Part one is about why characters ripped. The model has no memory, so anything you leave out, it invents. They will fix that with the character Bible. One page in plain nouns, so nobody has to guess. A couple of years back, I made an early explainer with Anika in it. In shot one, she wore the cream turtleneck. Perfect. And shot two was still fine. By shot three, she was in a gray hoodie with a draw string, which I certainly hadn't asked for. I hadn't really asked for anything to be fair. I'd written cream turtleneeck once in the first prompt and then got lazy and wrote things like Same Woman now typing. The model didn't know what Same meant, and it never does. That's the first thing to understand, and it's the thing nobody tells beginners. Every time you press generate, the model starts from scratch. It has no memory of the last image. None. It's not looking at your previous picture and thinking, that woman again. It reads only the words in front of it right now and builds a person from those words alone. We'll spend about 16 minutes here because once this clicks, everything else is common sense. So what happens to everything you didn't write down? The model decides it for you, fresh every time. Let's do the arithmetic because it's a bit shocking. Say you leave out just ten details, hair length, hair color, eye color, earrings, sleeve length, trouser color, shoes, skin tone, age, the shape of the smile, and say, each one has only two likely versions. Two times two is four times two is eight. And by the tenth detail, you're at 1024. That's how many different people your vague prompt describes. You get one of them at random every single time. That's why your three Project one images looked like cousins. The bit that bothered you most, the face or the outfit or the hair was one of those unwritten details being rolled again. Think of it like ordering a sandwich over a bad phone line. If you only say cheese sandwich, you might get cheddar on white or Brie on sourdough and both are technically correct. The person making it isn't being difficult. They're filling the gaps you left. The model is that sandwich maker every single time, and it never remembers your last order. If you didn't write it down, the model will make it up. The character Bible. The fix is a document I call the character Bible. Animators have done this for nearly a century. They call the reference sheet a model sheet, and a drawing that matches it is on model. A drawing that doesn't is off model, and in an animation studio, that's a word that gets you a redo. We're borrowing their idea. One page that says exactly who this person is, so nobody human or machine gets to guess. Let me walk you through Anikas section by section, and you write yours alongside. The page runs top to bottom in this order. Face first, then hair, then the signature item, then wardrobe, then props, then palette, then style, then the poses. Picture a passport page that someone very fussy has filled in. That's the feel. The face comes first, and I don't mean pretty or striking. Those words mean nothing to a model because they mean something different to everyone who reads them. Write things you could point at instead. Anika has warm tanned skin, warm brown eyes, defined brows, and a soft focused smile. Think of a police sketch artist who never asked the witness if the man was handsome but asks about the nose, the brows, and the jaw. Be that sketch artist. Hair next, and be specific about shape as well as color. Black curls worn in a high messy bun with loose tendrils framing the face. Tendrils is a lovely word here, by the way. It just means those little strands that escape and hang down by the cheeks. Write them in, or the model will tidy them away. Now the signature item, and this one matters more than anything else on the page. It's one small distinctive thing the character always wears. For Anika, it's small gold hoop earrings. For another character of mine, Emma, it's gold hoops with a tiny white five petal flower charm. One specific accessory gives the model something hard to hold onto, and it gives you something fast to check. If the earrings are missing, you know the image is off before you've even looked at the face. Then wardrobe. Write one hero outfit, the one she wears by default, and write it in pieces. Hers is a cream ribbed knit turtleneck with sleeves pushed to the forearms, high waisted olive green trousers, a black belt with a gold buckle and black loafers. Notice the word ribbed kit, and notice that I said where the sleeves sit. Each of those is a coin the model would otherwise toss. If you want other outfits later, list them underneath as rotations, so the outfit changes on purpose and never by accident. Here's a quick example of the difference that makes. My early line for Anika said smart casual top and trousers. Across six images, I got a blouse, a polo shirt, a cardigan, and two different jumpers. When I changed it to cream ribbed knit turtleneck with sleeves pushed to the forearms, I got the same top six times out of six. Same character, same tool, just better nouns. Props are the things she holds or uses. Ankas are a space gray laptop, a black mat mug, a rust spiral notebook, and black over ear headphones. Palette is her colors named properly, cream, black, rust orange, olive, beige, gray. Blue isn't a color on a Bible page, by the way. Blue is 1,000 colors. Navy is a color. Then a style line, one sentence that says how she's rendered. Hers says, semi realistic painterly three D, warm and cinematic. And last, a short pose list in a sentence which we'll use in the main project. If you've been doing this awhile, here's the bit I bet you're getting wrong. You're probably describing your character with a reference to a famous face or a famous film look. Don't. It pulls the model towards someone else's identity, and it's not yours to borrow anyway. The sharper move is to keep a Bible with no names in it at all, only nouns, colors, and measurements. It's more boring to write. It's also the version that holds across 50 images and three different tools. A few things go wrong here, and I've done all of them. The first is writing adjectives instead of nouns. My first Bible for Anka said elegant and warm and professional, which are lovely words and completely useless because the model gave me three different elegant women. The second is a Bible that's too long. If your page runs to two pages, you've written a novel and half of it won't fit in a prompt. The third is forgetting the boring bits like shoes because shoes are where drift loves to hide. Now it's your turn. So take about 7 minutes and write your character's Bible in your text file. Work down the page through face, hair, signature item, wardrobe, props, palette, style, and poses using plain nouns and short lines. If you get stuck on the face, describe it the way you describe a friend to a taxi driver collecting them from a busy station. Before we move on, ask yourself two things. Could a stranger draw your character from this page without asking you a single question? And does your page have one signature item a person would spot in a crowd? If the answer to either is no, Adeline, then we're building the room. Write her down once in nouns, and the model has nothing left to guess. If you didn't write it down, the model will make it up. 6. Part 2: The_Frozen_Set_and_the_Locked_Block: In part two we lock the room and lock the words, the frozen set keeps every object count on light in place. The locked block means you change one thing at a time and nothing else moves. The first proper Anika video had a little plant on the left of her desk. Small green thing, speckled white pot. In the first shot, it had seven leaves. I didn't count them then. I counted them later after a reviewer on the client side asked quite politely, whether the plant was meant to be growing. In shot two, it had 11. In shot four, the pen cup had three pens. In shot five, it had a fist full. Anika was fine by then. Her room was having a breakdown. That's the other half of the problem, and it's the half most people never think about. You can lock the character perfectly and still lose the shot because the background quietly rebuilds itself every time. Your viewer might not say the plant grew. They'll just feel that something's off. This is the core of today, so we'll take about 24 minutes over it. Build the room once. The fix is a habit I call the frozen set. Treat your scene like a real film set that's been built, dressed, and then locked. Every object keeps its place, its count, and its direction. The light never moves. The only things allowed to move are the character and the camera. Think of a stage crew putting tape marks on the floor, so the sofa lands in the same spot every night. We're doing the tape marks in words. Let me draw Anika set for you, the way I'd sketch it on a whiteboard. Picture a long walnut desk running across the frame left to right, like a shelf in a shop window. Now walk your eye along it from the left. First, a small green plant in a speckled white pot. Next to it, a terrazzo pen cup holding five black and gold pens. Then a black mat mug sitting on a round cork coaster. Front and center, a rust spiral notebook, right in the middle, a space gray laptop, and at the far right end, a pair of black over ear headphones. Now lift your eye above the desk. Behind it is a cream wall. A tall window sits at the far left, and that's where the light comes from. Up at the top right, floating wood shelves with a trailing pathos and a black vase. At the back right, a mushroom shaped lamp glowing warm and one tall, large leaf plant at the very right edge of the frame. That's the whole room, and you could rebuild it from memory now. Grab a pen and sketch that in your margin. 10 seconds, no ruler. You'll remember it far better than if I'd shown you a picture. Notice what I did there. I went left to right. I gave a count wherever there was more than one of something, and I named the light by where it comes from. The line for this set says golden hour sun from the window on the left, Rim light on her hair and shoulders. Soft haze, shallow depth of field. Rim light just means the bright edge of light that outlines her from behind. Depth of field is how much of the picture is sharp. Shallow means she's sharp and the back wall goes soft. One opinion, and I hold it firmly. Keep the set small. Every object you add is another object that can drift. I've seen gorgeous proms with 40 props and two bookshelves and a cat, and not one of those rooms survived a second generation. Six or seven things on the desk is plenty. Minimal isn't a style choice here. It's insurance. Build the room once. Then only the camera is allowed to move. The Locked Block. Now, we put the Bible and the set together into the thing you'll actually paste. I call it the Locked Block. It's one block of text always in the same order pasted word for word into every generation, and it has exactly one open slot, a gap where the only thing that changes goes. Here's the order and say it with me because it's worth memorizing. Style comes first, then identity, which is the face, hair, and signature item from your Bible. After that comes wardrobe, then the frozen set, read left to right with counts, then lighting, then the camera and the frame. The open slot always goes last. Style identity, wardrobe, set, light, camera, slot. Seven pieces in the same order every single time. Why the order? Because consistency loves boring. If the earrings are mentioned third in one prompt and 12th in the next, you've changed the prompt, even if every word is the same. And the model tends to treat whatever comes early as more important. So the things that must never change go near the top, and the thing that should change sits at the bottom, all on its own. Picture a pilot's preflight checklist. The pilot has flown that plane 1,000 times and still reads the list in the same order out loud before every takeoff, not because they've forgotten, because the order itself catches the mistake. Your locked block works the same way, and the dull repeated order is exactly where its strength comes from. Let me read you a real one so you can hear it. Semi realistic painterly three D, warm cinematic look. A woman with warm tanned skin, warm brown eyes, defined brows and a soft focus smile. Black curls in a high messy bun with loose tendrils framing her face, small gold hoop earrings, Cream ribbed knit turtleneck with sleeves pushed to the forearms, high waisted olive green trousers, black belt with gold buckle. Then the set, red out left to right exactly as I drew it for you. Then the light line, then eye level mid shot 16 by nine. And then the slot. The slot for the first image says, standing behind the desk, both hands on the laptop keyboard, eyes on the screen. That's it. For the next image, I copy the entire block, paste it, and change only the slot. Now it says, holding the black mug in both hands near her chin, looking at the camera. Nothing else moves, not a comma. That's the one slot rule, and I'm a bit of a bore about it. Change one thing per generation. If you change the pose and the angle and the expression all at once, and the face drifts, you have no idea which change broke it. Change one. And when something goes wrong, you know exactly what did it. It's the same as debugging code. You never change three lines and then wonder which one fixed the bug, paste the block, change one word, never two. If you've done this a while, here's your sharper beat. Once you have one image you're truly happy with, make it your anchor render. That's the single approved picture every other image gets judged against. If your tool lets you attach a reference image, attach the anchor render every time, along with the full locked block. Don't let the reference replace the words, though. The reference pulls the look. The words pin the details. People who rely on the reference alone are the ones who end up with the right face in the wrong jumper. A few things go wrong here. The big one is editing the block in place. You tweak a word in the prompt box, generate, tweak again, and 20 minutes later, your block has quietly mutated and you can't remember what the original said. Fix that by keeping the master copy in your text file and only ever pasting from it. The prompt box is a scratchpad. The file is the law. The second is a slot that sneak set details in. Someone writes in the slot. Now she's by the window drinking tea. Suddenly, there's a tea cup the set never had, and the windows moved. Keep the slot about the character. Pose, expression, angle. If you want her somewhere else, that's a new frozen set, written properly, not a sneaky line in the slot. The third is the reverse angle. When the camera swings around to look the other way, it sees a different wall. Students expect the model to work that out. It won't. A reverse angle needs its own frozen description written once, kept forever. Same rules. For Anika, I have a front angle set and a reverse angle set, and each one is locked. Your turn and give this a solid 10 minutes. Write your frozen set in your file, left to right with counts and a light line. Then build your locked block in the seven piece order with the slot sitting at the bottom. Generate one image, the one. If it's close to what you wanted, that's your anchor render. If it isn't, check the order before you change a single word. Before we move on, two questions. Is there anything in your slot that isn't about the character? And if I asked you right now, how many pens are on your desk, could you answer from your file without looking at the picture? If you can, you've got a frozen set. If you can't and count. Lock the room and lock the words, then change one thing at a time, paste the block, change one word, never two. 7. Part 3: The_Seven_Leaf_Test_and_Fixing_Drift: P three is where we stop trusting our eyes. You eye lies when it's happy, so we count things before we like them. That's a seven leaf test. Then we name the drift and fix it one change at a time. Here's a confession. Even with a perfect Bible and a locked set, I still get images that are slightly wrong. Last month, I had an Anika shot I loved on site with gorgeous light and a lovely smile, and I nearly sent it straight to the client. Then I counted the pens and found four when the set clearly says five, my eye had been so pleased with the smile that it didn't bother checking the disk. That's the trap in this last part. Your eye lies when it's happy. So we need a check that doesn't rely on feelings. I call it the Seven-Leaf Test. After that plant that grew from seven leaves to 11. After every generation, before you decide you like it, you count things. We'll spend about 16 minutes here, and this is the part that turns lucky images into a reliable set. The test runs the same way every time, and it takes under a minute. Start with the signature item and check that both gold hoops are there, then the hair, which should be the high bun with tendrils, then the wardrobe, piece by piece from the ribbed turtleneck with pushed up sleeves down to the olive trousers and the belt. Then walk the set left to right with counts, plant, five pens, mug on its coaster with the handle facing the same way, notebook laptop, headphones. Finish with the light and make sure the sun is still coming from the left. If a single count is wrong, the image fails. It doesn't nearly pass. It fails and you go again. That sounds harsh, but it's actually freeing because it takes the mood out of the decision. You stop arguing with yourself about whether something's close enough. The counts decide and you move on. Count the leaves. The leaves don't lie. Which drift have you got? When an image fails, the next job is to name the problem because each kind has its own fix. I sort them into four. Face drift is when the person changes, a different nose, older, younger, a different smile. Wardrobe drift is when the clothes change, the hoodie problem. Set drift is when the room changes, the growing plant, and light drift is when the sun moves, the time of day shifts, or the whole picture goes cold and blue. Think of it like a mechanic listening to an engine. A rattle, a squeal, and a knock all mean the cars unhappy, but nobody fixes all three with the same spanner. Go back to the sentence you wrote at the top of your file after Project one. Which of the four was it? Most of you will find it was face drift or wardrobe drift. Now you've got a name for it, and a name is half a fix. For face drift, check your identity lines first. Nine times out of ten, there's an adjective in there doing a nouns job. Swap it for something you could point at. Then check whether the face sits early in the block. If it's buried under the set description, move it up. If your tool takes a reference image, this is where the anchor render earns its keep. Wardrobe drift, look for a missing piece because the model has usually invented the bit you didn't name. It's often the shoes or the sleeves, so name them. And check the slot hasn't brought in a word like cozy or casual, because the model hears those and reaches for a hoodie every time. I'm not joking about this. Cozy is a hoodie magnet. For set drift, count your set description against the picture. Missing counts are the usual culprit. Write five pens, not some pens. Write the mug handle facing right. And if the set keeps slipping, cut an object. Remember, fewer props, fewer things to drift. Say your mug keeps hopping from the left of the laptop to the right. Before you touch anything else, read your set line out loud. Nine times out of ten, you hear it straightaway. You wrote a mug near the laptop instead of a black mat mug on a round coke coaster to the left of the laptop, handle facing right. Vague words give the model permission to wander. For light drift, check whether your slot mentions time or mood. A line like a late night at work will drag the sun straight out of the window. Your line is part of the set, so it never changes. If you want a night scene, that's a different set written and locked on its own. If you've been at this a while, here's the bit I bet you're getting wrong. Expressions, a big laugh, a shout, a crying face. They pull the face off model faster than anything else, because the model rebuilds the whole face to fit the emotion. Dial them in small steps. Start with a soft smile, then a wide smile, then laughing with eyes closed. Three small steps, each one tested, beats one big leap that hands you a stranger. A couple of other things go wrong here, too. One is fixing two drifts at once, which breaks the same rule as the slot. So change one thing, regenerate and test. Another is falling in love with a failed image. I know. The light is so good. But a failed image in a set of six makes the other five look wrong too. So put it in a folder colled nearly and let it go. Your hands on stretch is about 6 minutes. Take the anchor render you made in part two. Change the slot to a new pose, generate, and run the full seven leaf test out loud. In a whisper, if you're in a cafe. If it fails, name the drift and make one fix. Then generate again. Before the big project, ask yourself this. When your last image failed, could you name which of the four drifts it was within 5 seconds? And did you make exactly one change before trying again? If both answers are yes, you're ready to work on your own. Every image gets counted before it gets liked. Count the leaves. The leaves don't lie. 8. Main Project: Six_Poses_One_Person: Right. This is where I stop talking you through it and you work alone. We've got about 20 minutes. Here's the brief in one sentence. Make six on model images of your character in six different poses inside one frozen set that pass the Seven-Leaf Test side by side. The six poses are these, and I'll say them as a sentence so you can copy it into your Bible. Standing neutral with one hand in a pocket, arms crossed, confident, thinking with a hand on the chin, on a call wearing a headset, walking, mid stride and working at the desk, hands on the laptop. That's a pose sheet. The same thing an animation studio would pin above a designer's desk. Every bit of this uses what we've already done. Part one gave you the Bible, so the identity and wardrobe lines are written. Part two gave you the frozen set and the locked block. So the room and the order are written. Part three gave you the seven leaf test and the four drifts. So you know how to judge each image and what to do when one fails. Nothing new today. You've got every tool you need. Here's how I'd spend the time. Spend the first 3 minutes reading your Bible and your locked block once more slowly, and fixing anything vague. Then give yourself about 2.5 minutes per pose. Paste the block, change the slot to the pose, generate, run the test. If it passes, it goes in the folder. If it fails, name the drift, make one fix, try again. Leave the last 2 minutes for the side by side. For example, here's what my desk pose slot looked like on Anika sheet. Sitting at the desk, both hands on the laptop keyboard, eyes on the screen, soft focused smile. For the next pose, I pasted the whole block again and changed only that line. Now it read standing beside the desk, arms crossed, looking at the camera with a confidence smile. Everything above the slot stayed exactly as it was down to the last comma. Two of the poses take a little thought, so let me warn you. Walking and the headset shot both pull the character away from the desk. For walking, keep the set behind her and say she's walking past the desk so the room stays in the frame. For the headset, add the black over ear headset to your props in the Bible first before you generate. Don't let the slot invent it. That way, it looks the same every time she's on a call. Now, make it yours. You choose the character, the style, and the room, flat two D, painterly three D, or photo reel. All of them work. If you want a harder version, do the six poses from two camera angles, front and reverse with a frozen set for each. That's 12 images, and it's the version I'd set an intermediate student. Here's where people stall and what to do. The first stall is the third or fourth pose where one stubborn image keeps failing. Don't regenerate it 20 times. After three fails, stop, name the drift, and change one line in the block itself, not the slot. The second stall is the walking shot losing the set. Pull the camera slightly wider in the slot. Mid shot to medium wide. The third stall is tiredness. By pose five, you'll be tempted to skip the test. That's exactly when the pen cup grows a sixth pen. How do you know you're done? Line the six up in a row. If you're still learning, you'll see the same character, but a couple of details wandering, a missing earring, a mug in the wrong spot, that's fine. That's progress. If you're getting there, every count passes, but the light shifts a little between shots. If you've got it, a stranger could look at the six and tell you without asking that they're one person in one room on one afternoon. And here's the arithmetic I promised. Six poses, three camera angles, three expressions. Six times three is 18 times three is 54, 54 images, all from one page. Now, go to the folder and pull out your three Project one images. Put them next to your six. Look at both rows for a moment. That first row is 10 minutes of guessing. The second is one page of decisions. Same, same tool, 100 minutes apart. If one of your six is still a bit wonky, that's fine. A wonky one in a row that holds together is miles ahead of three strangers. You don't need better luck. You need fewer guesses, and you just removed most of them. 9. Conclusion_Six_Poses_Not_Six_Cousins: Let's take stock. At the start, I promised you six poses of one person, not six cousins. You wrote a Character Bible, so the model stopped guessing who she is. You built a frozen set, so the room stopped rearranging itself. You wrote a locked block with one open slot, so you could change a pose without breaking a face, and you learned the seven leaf test, so you judge images by counting rather than hoping. If you only remember one thing, remember that the model has no memory. Every image is a first meeting. So you bring the introduction with you every time word for word. That's all a Character Bible really is a very good introduction. Here are three things to make next, each one harder. First, take your six and add three expressions to the thinking pose. Soft smile, wide smile, laughing in small steps. Then write a reverse angle frozen set and redo the desk pose from behind her shoulder, looking toward the window, and then the big one. Put your character into a short story of eight shots with the room, the light, and the face holding from first frame to last. That last one is exactly what I do for client videos, and it's where all of this pays off. To keep improving on your own, keep one habit. Every time an image fails, write a single line in your text file saying which drift it was and what fixed it. In a month, you'll have your own little repair manual written in your own words for your own characters. Thanks for spending the hundred minutes with me. Go count some leaves. How I built this course? I built this class around the frustration I hear most from people starting out with image tools. The character won't stay the same. I assumed you generated a few images before but had no system. I also assumed you might be using any tool at all. So I kept every move tool-agnostic and taught it through Anika, a real character from my client's video work. If your learners are already comfortable with prompts, shorten Part one and spend the time on reverse angles and expression steps in Part three. Thanks for learning. Keep supporting me. See you in the next course. That's the course. You came in with guesses and you are leaving with a system. Thank you for learning with me overall this time. This is just a start. There's plenty more we will learn together until then go count some leaves. Thank you.