I've been getting a lot of questions about how I animated my Doodles PFP, so here's the full breakdown. Way more detailed than what I posted on X. I'll cover every step from scratch, including the exact prompts and tools. Everything you need to replicate this for your own PFP is in the downloads at the bottom.
The workflow in one line
Character sheet → style references → generate your still image in ChatGPT → structure the prompt with Claude → generate the video in Higgsfield on Seedance 2.5 → edit and sequence the final cut. The downloads at the bottom give you the two tools that make it work.
The finished animation. This is what the whole process produces.
What You'll Need
Step 1
The character sheet is the foundation. Every image and video you generate will be anchored to this, so getting it right upfront saves you hours of iteration later.
Create a simple 3-panel reference sheet: Panel 1 is the front body with no head. Show the full outfit from neck down, all accessories, clothing details, shoes. Cut off at the neck. No head. Panel 2 is the back view, full body. Complete character from behind including the back of their head, hat, or hair. Panel 3 is a close-up bust shot from the chest up, front-facing, centered. This is your face reference panel and the only place the face appears.
The headless front panel might feel counterintuitive, but it's intentional. When the AI sees the full character with a head in one panel AND another full-body shot with the head, it tries to reconcile both and generates inconsistently across runs. Separating them keeps it clean.
The most important thing I learned: only show the face from a single angle in that third panel. The moment you try showing multiple face angles (front and side, for example) the model starts shifting and your character looks slightly different every generation. One clean, forward-facing shot locks the character in.
To use the prompt below: paste it into ChatGPT, then attach two images. First the character sheet reference I provided in the downloads below, then your own Doodles PFP. The prompt handles the rest, converting your PFP into the same illustration style.
Key Tips
The 3-panel character sheet I used as my AI reference throughout the whole process.
Create a clean, professional 3-panel character reference sheet for the person in the provided profile picture, using the provided character sheet only as a style, layout, and presentation reference. REFERENCE ROLES If two images are provided, use them like this: Image A = the character sheet reference. Image B = the profile picture / PFP reference. Image A is for reference only. Use Image A as the source of truth for: • Overall illustration style • Shape language • Simplification level • Line quality • Line thickness • Coloring style • Shading style • Character sheet layout • Panel composition • Background treatment • Framing • Overall presentation Do not copy the identity of the character shown in Image A. Do not recreate the character from Image A. Do not use Image A as the identity reference. Image B is the actual character to be turned into a character sheet. Use Image B as the source of truth for: • Facial identity • Face shape • Hair shape • Hair color • Skin tone • Facial features • Visible accessories • Visible clothing • Visible neckline and upper-body design • Overall likeness of the person or character in the PFP The final result should look like the subject from Image B redrawn as a character in the exact same visual style and character-sheet format as Image A. CORE GOAL Turn the subject in Image B into a stylized character sheet that feels like it belongs in the same world, same art style, and same presentation system as Image A. The final image should feel like Image B has been converted into a character reference sheet using Image A as the style template. Do not create the original character from Image A again. Do not mix the identity of Image A with the identity of Image B. If any instruction conflicts with the visual references, prioritize the references based on the role definitions above. STYLE TRANSFER LOCK Match Image A as closely as possible in: • Illustration style • Flatness or rendering style • Proportions style • Cartooning approach • Facial simplification • Eye style • Mouth style • Hair rendering style • Clothing rendering style • Outline thickness • Line smoothness • Color handling • Shading level • Background simplicity • Overall visual personality The new character should clearly look like the subject from Image B, but drawn in the same style as Image A. Do not make the result more realistic than Image A. Do not make the result more detailed than Image A. Do not make the result more painterly, more anime, or more 3D than Image A. IDENTITY LOCK The character in the final sheet must clearly represent the subject from Image B. Preserve from Image B: • Hair silhouette • Hair part and shape • Skin tone • Face shape • Facial spacing • Visible expression style • Visible accessories • Visible clothing colors and clothing type when clear If Image B contains only the head or upper torso, extrapolate the missing lower-body details conservatively and logically in the same style, based on the visible outfit. Keep the invented lower-body design simple, clean, and consistent with the visible upper-body clothing from Image B. Do not invent random details that are unsupported by Image B. If a clothing detail is clearly visible in Image B, preserve it. If a detail is not visible in Image B, keep the design minimal and stylistically consistent with the look of Image A. CANVAS + LAYOUT Create one wide image divided into 3 vertical panels side by side. All 3 panels must: • Be equal in width • Use the same plain neutral studio background • Be separated by thin, subtle vertical divider lines • Use consistent lighting • Use consistent line weight • Use consistent rendering and shading • Maintain identical character proportions • Match the exact same art style • Feel like one cohesive professional character reference sheet Keep the presentation simple, clean, and uncluttered. Do not add text, titles, labels, measurements, arrows, logos, annotations, or decorative graphics. PANEL 1: FRONT BODY REFERENCE, HEAD CROPPED Show the character standing directly front-facing in a neutral, relaxed pose. Frame the character from just below the head down to the feet. The head should be completely cropped out of the frame. Clearly show the character's: • Upper-body clothing • Lower-body clothing • Clothing construction • Accessories • Hands • Arms • Torso proportions • Leg proportions • Footwear • Any visible body or outfit details Arms should rest naturally at the sides. Keep the pose simple, symmetrical, and useful as a production reference. The feet should sit on the same baseline as the character in Panel 2. PANEL 2: FULL-BODY BACK VIEW Show the exact same character directly from behind. The entire character should be visible from head to feet. Use a neutral standing pose with the arms relaxed naturally at the sides. Match the exact scale, body proportions, and height established in Panel 1. Clearly show the design from the back, including: • Back of the head • Hair or head details • Back of clothing • Back of accessories • Sleeves • Lower-body clothing • Footwear from behind • Any design details that would naturally be visible from this angle Only show elements that would realistically be visible from the back. The character should occupy approximately 85 to 90 percent of the panel height. The feet should sit on the same baseline as Panel 1. PANEL 3: HEAD + UPPER BODY CLOSE-UP Create a large close-up portrait of the exact same character. Frame approximately from the upper chest or shoulders to slightly above the head. Use a slight three-quarter angle unless a direct front-facing angle better represents the subject from Image B. This panel should clearly show the character's most important identity and design details. Accurately reproduce: • Head shape • Face shape • Facial features • Hair • Accessories • Jewelry if present • Neckline • Upper clothing • Any unique character-defining details from Image B Render these features using the exact style language established by Image A. CHARACTER CONSISTENCY LOCK The character must clearly be the exact same individual in every panel. Keep completely consistent: • Body proportions • Height proportions • Head proportions • Shoulder width • Arm length • Leg length • Hand design • Skin tone • Hair color • Hair shape • Clothing colors • Clothing construction • Accessories • Footwear • Facial proportions • Character-specific design features • Art style • Line thickness • Shading style • Color treatment • Level of detail Do not change the outfit between panels. Do not create alternate versions of the character. Do not add props unless they are clearly part of the subject's design. Do not create multiple poses inside a single panel. Do not exaggerate or simplify the subject beyond the style language already established by Image A. FINAL GOAL The final image should look like a professional character reference sheet of the subject in Image B, drawn in the exact same style, structure, and presentation as Image A. In simple terms: Image A tells you how to draw it. Image B tells you who to draw.
Step 2
Before generating a single image, spend time pulling reference shots from your project's official social media. The goal isn't to copy. It's to extract the visual DNA of the art style.
For Doodles, I went through their Instagram and saved 4-6 strong images. I was looking for specific things: the lighting quality (soft, painterly, warm golden tones), the world-building aesthetic (tropical, beachside, slightly nostalgic), the color palette (pastels mixed with saturated accents), and the overall mood (chill, aspirational, unhurried).
Feed these reference images into ChatGPT alongside your character sheet. This gives the model a concrete visual world to work from instead of generating generic-looking AI output. The more specific the style information, the more specific and on-brand the result.
Key insight: references aren't just inspiration. They're data. Treat them that way.
Key Tips
The reference images I pulled from Doodles' Instagram: lighting, world-building, color palette, and mood.
Step 3
This is where iteration lives. I didn't nail it on the first try. I went back and forth probably 8-10 times before landing on the final image. That's normal. Budget for it.
My process: start with a clear scene description ('My character sitting on a silver DeLorean on a Miami beach at golden hour, palm trees in background, Doodles art style'), then feed in the character sheet plus reference images and study the output. Identify what's off (wrong lighting, character drifting from the sheet, background too generic) then iterate on the specific issue, not everything at once. Bad iteration: 'Make it better and more like the reference.' Better iteration: 'Keep everything exactly the same. Only move the floating eye slightly closer to the face so its spacing matches the character sheet.' That specificity is how you actually move the image forward.
What made the biggest difference: being extremely specific about lighting ('soft golden hour light from the left, warm rim glow on the character's silhouette'), calling out the art style by its source ('flat illustration with painterly textures, the visual world of Doodles circa 2022'), and when a reference image was close in composition to what I wanted, saying 'use this framing but replace the subject with my character from the character sheet.'
I used one of their shots as a composition reference with the same general framing and character placement, and iterated until the result was clearly its own thing. Starting from a composition you like and adapting it through multiple rounds of iteration is how you end up with something original. By the time I was done, the final composition had evolved into something completely specific to the scene I was trying to create.
This image is everything. It becomes your starting reference, and all the video generation builds on top of it.
Key Tips
Step 4
This is where Claude enters the workflow. Honestly, it's the step that changed the quality of my outputs the most.
I use a Claude skill called cinema-director-v3 (download link below). Once you add it to Claude, you describe your scene in plain language and it returns a fully structured video generation prompt: shot type, camera movement, subject motion, ambient animation, lighting spec, art style, and duration. If you haven't used Claude before, it's a free AI tool from Anthropic. The download file includes setup instructions and it takes about two minutes to get running.
Instead of typing something vague like 'character sitting on a car at the beach,' you describe it conversationally and the skill translates it into cinematic language that AI video models actually respond to. Think detailed camera directions, how each element in the frame moves, light behavior, texture specs.
Example: describe your scene as 'my Doodles character sitting relaxed on a DeLorean on a Miami beach at sunset, palm trees behind them, calm ocean in the background' and cinema-director-v3 turns that into a prompt specifying shot type, whether the camera moves, how the chains catch light, how the palm trees sway, the atmospheric haze. Everything the video model needs to generate a coherent, cinematic clip.
This structured prompt approach is what gets you consistent results instead of random outputs. The model has clear instructions for every frame instead of guessing.
Key Tips
💡cinema-director-v3 is in the downloads below. Drop it into Claude and it's available in every conversation going forward. No setup beyond that. Never installed a Claude skill before?Here's how to do it →
Step 5
Take the structured prompt into Higgsfield and run it on Seedance 2.5, using the image you generated in Step 3 as the reference image.
A few settings that matter: use your generated image as the reference, not the character sheet. That image already has all your style and scene baked in. It's a much better starting point. Seedance 2.5 is specifically strong at character consistency across frames, which is exactly what you need for PFP animation. The base model drifts more.
Start with 4-6 second clips. Shorter clips are easier to get right before pushing to longer durations. Once you have one clean take, you can run the same character through different scenes and start building a full video from multiple clips.
Key Tips
Step 6
Once I had a few generations I liked, I treated them like normal footage. Naturally, as an editor, I wanted to bring each scene to life with real SFX and audio. I transitioned each scene with a swipe and a whoosh SFX.
That's what separates this from just posting a raw AI generation. The edit is where everything comes together into something that actually feels intentional.
Key Tips
The Full Workflow
Free Resources
All files in one place: the Claude skill, character sheet prompt, and character reference image. One click, then grab what you need.
The actual Claude skill file I use. Not a simplified version. Drop it into Claude and describe any scene in plain language. Setup instructions are inside.
The 3-panel character sheet used as Image A in the ChatGPT prompt. Upload this alongside your PFP when you run the prompt so ChatGPT knows the style and layout to follow.
The exact ChatGPT prompt I use to turn any PFP into a clean 3-panel character sheet. Upload the provided character sheet reference alongside your PFP, paste in the prompt, and ChatGPT handles the rest.
The full body reference image of my Doodles character. Use it as a starting point if you want to practice the workflow with the same character before using your own.
FAQ
Do I need Claude Code to use cinema-director-v3?
No. You can install the skill directly inside Claude.ai. Go to Customize > Skills, upload the skill file, and enable it. You may also need to enable Code execution and file creation under Settings > Capabilities. No Claude Code required.
Does this workflow only work for Doodles?
It can work for other projects, but there are a couple of extra steps. You'll need to find a full-body reference image of your NFT, since most collections don't provide one. You may also need to adjust the character sheet prompt to match the specific art style of your project. The overall workflow is the same, but the setup takes a little more manual work when you're starting from a project that isn't Doodles.
What if my generated video doesn't look right on the first try?
That's normal. Expect to generate 3-5 variations before landing on a clean take. If the character is drifting, your reference image is your biggest lever. If the motion looks wrong, revise the prompt through cinema-director-v3 and be more specific about what moves and what stays still. The prompt is almost always the issue, not the model.
If you make something with this workflow, I want to see it. Drop a reply or tag me.
@itspatj_Work With Us
