Video Character Design
Use this skill to create a reusable character asset package for video planning, storyboard generation, and image-to-video generation.
The package has two files:
- — the character design specification.
- — the character design sheet image generated only after the user confirms the .
The user-facing
may be written in the user's preferred local language unless they request English. Keep file names simple, lowercase, and filesystem-safe.
Image Generation Requirement
Character sheet images must be generated by an image-generation model. Do not create character artwork with code, SVG, HTML/CSS, canvas, diagrams, placeholder drawings, or deterministic shape scripts. Those methods are acceptable only for non-visual metadata, file operations, or optional post-processing after the model-generated character sheet exists.
Use the built-in
tool by default, or an approved project image-generation workflow when explicitly required. After generation, save or copy the model-generated raster output to
characters/{character-name}.png
. If the model output cannot be saved directly, locate the generated raster asset and move/copy it into the required path; do not recreate it programmatically.
Storyboard Reference Discovery
Before creating the character spec or generating the character sheet image, actively look for existing storyboard imagery that can anchor the character design.
Check both:
- The current conversation/context for any generated storyboard images, scene boards, uploaded storyboard references, or previously created assets.
- The current project's storyboard directory, especially , for corresponding scene images such as , , or any file the user mentions.
If a relevant storyboard image exists, treat it as a required visual reference for both extraction and image generation. Include its path in the
under
, describe which character is being extracted from it, and pass the image as a reference input when generating
. This matters because the character sheet must preserve continuity with already generated storyboard frames instead of redesigning the person from text alone.
If multiple storyboard images show the same character, use all relevant images as references when practical, and note any differences or assumptions in the
.
Core Rule: Spec First, Image Second
Always create the
design spec first.
Do not generate the
in the same step unless the user has already explicitly approved an existing character spec or says to proceed without confirmation.
Default sequence:
- Extract or design the character.
- Save .
- Ask the user to confirm or revise the design.
- After approval, generate .
- Save the generated image beside the .
This confirmation step matters because character images are expensive to iterate after the identity, wardrobe, and sheet layout are wrong.
Output Location
Use a project-local character directory when one exists. If no convention exists, create:
Save outputs as:
text
characters/{character-name}.md
characters/{character-name}.png
If the user explicitly requests another directory, use that directory.
Examples:
characters/campus-girl.md
characters/campus-girl.png
characters/travel-boyfriend.md
characters/blue-team-runner.png
Do not overwrite existing files unless the user requested replacement. For alternatives, use suffixes such as
,
, or
.
Input Modes
Support two input modes.
Mode A: Extract From Reference
Use this when the user points to an image, storyboard, video still, uploaded reference, or existing generated asset.
Also use this mode when storyboard imagery is discovered in the current context or project
directory, even if the user only asks generally to make a character sheet for a video project.
Extract observable traits:
- Apparent age range and casting type.
- Face shape, hairstyle, skin tone, expression, and distinguishing features.
- Body proportions and posture.
- Wardrobe, shoes, accessories, bag, jewelry, props.
- Color palette and materials.
- Clothing material details: weave, knit, sheen, thickness, drape, seams, cuffs, stitching, wrinkles, weathering, and how the fabric reacts to studio light.
- Scene context, lighting, and camera feel.
Be honest about uncertainty. If a detail is unclear, mark it as an assumption instead of inventing a precise fact.
Mode B: Design From Scratch
Use this when the user describes a character without a reference image.
Ask only for missing details that materially change the character:
- Character role in the video.
- Age range or life stage.
- Gender presentation if relevant.
- Setting and genre.
- Wardrobe direction.
- Visual tone: realistic, commercial, documentary, stylized, animation, etc.
If the user wants speed, make reasonable assumptions and list them in the
.
Character Name
Derive a concise slug from the role when the user does not provide a name.
Rules:
- Lowercase.
- Use hyphens.
- ASCII only.
- Avoid spaces and punctuation.
- Keep it stable across revisions.
Examples:
- "校园广告里的女生" ->
- "男主,旅行情侣里的男生" ->
- "basketball team captain" ->
Template
Use this structure for the design spec. Translate headings if the user prefers another language.
markdown
# [Character Display Name]
## Source
- Input mode: [Extracted from reference / Designed from scratch]
- Reference file(s): [paths or "none"]
- Project context: [video/storyboard/ad/scene]
## Design Intent
[One short paragraph explaining the character's role, personality, and visual function in the video.]
## Identity & Casting
- Apparent age:
- Gender presentation:
- Ethnicity / cultural context:
- Build and height impression:
- Posture and movement quality:
## Face & Hair
- Face shape:
- Skin tone and texture:
- Eyes:
- Brows:
- Nose:
- Mouth:
- Hairstyle:
- Distinguishing details:
## Wardrobe
- Top:
- Bottom:
- Shoes:
- Socks:
- Outerwear:
- Accessories:
- Bag / carried items:
- Color palette:
- Fabric and fit:
- Clothing material details:
## Props
- Primary props:
- Secondary props:
- Prop consistency rules:
## Character Sheet Layout
- Canvas: 16:9
- Background: photography studio gray, plain, no lines or patterns
- Row 1: full-body turnaround, about 60% height
- Row 2: face or detail close-up turnaround, about 40% height
- Views: left side, front, back, right side
- Pose: natural standing posture, professional studio fitting-test look
## Image Generation Prompt
[A complete prompt ready to generate the character sheet after approval.]
## Negative Prompt
- No text
- No labels
- No grid lines
- No decorative background
- No extra logos
- No prop changes between views
- No wardrobe changes between views
- No hairstyle changes between views
- No face changes between views
- No distorted hands
- No extra fingers
- No plastic skin
- No cartoon or illustration style unless requested
## Continuity Rules
- [Specific rules for keeping this character consistent across future storyboards and videos.]
## Open Questions
- [Only include questions that are still worth confirming.]
Character Sheet Image Requirements
Default sheet format:
- 16:9 canvas.
- One single image.
- Two rows.
- Row 1 takes about 60% of the height.
- Row 2 takes about 40% of the height.
- Background is plain photography-studio gray.
- No text labels, no grid lines, no decorative pattern.
- Subtle natural gradient is allowed.
- Very faint character shadow is allowed, but it must not dominate.
Row 1:
- Full-body views from head to toe.
- Include left side, front, back, and right side views.
- Natural posture.
- Same scale across all four views.
- Professional studio fitting-test photography feel.
- Clothing materials should be carefully rendered when applicable: show fabric texture, thickness, seams, hems, cuffs, stitching, folds, wrinkles, subtle sheen, and realistic drape instead of treating the outfit as a flat color block.
Row 2:
- Face close-up views.
- Include left side, front, back, and right side views.
- Same identity, hairstyle, and expression logic as Row 1.
- If the character has no face, use detail close-ups instead: head, hands, outfit material, props, silhouette, mechanical details, or other important design features.
Continuity Requirements for Image Generation
When generating the
, explicitly emphasize:
- Same character in every view.
- Same face and facial proportions.
- Same age and body proportions.
- Same hairstyle from all angles.
- Same wardrobe and fit.
- Same clothing materials and construction details: fabric texture, seams, stitching, hems, cuffs, wrinkles, drape, and sheen should remain consistent across all views.
- Same shoes, socks, accessories, and props.
- Same color palette and material behavior.
- Same studio lighting and camera distance within each row.
For reference extraction, tell the image model to preserve the character from the reference while converting it into a clean studio character sheet.
For from-scratch design, tell the image model to follow the approved
exactly.
Recommended Image Prompt Structure
Use this structure after the user confirms the
:
text
Create a 16:9 character design sheet for [character display name].
Use the approved character spec exactly:
[brief identity, face, hair, body, wardrobe, clothing materials, props, palette]
Layout: two horizontal rows.
Top row occupies about 60% of the canvas: full-body turnaround views from head to toe, left side, front, back, right side.
Bottom row occupies about 40% of the canvas: face close-up turnaround views, left profile, front face, back of head, right profile.
Continuity is critical: same character identity in every view, same face, same hairstyle, same body proportions, same wardrobe, same clothing material details, same shoes, same accessories, same props, same color palette.
Render clothing materials carefully when applicable: visible fabric weave or knit, realistic thickness, seams, hems, cuffs, stitching, natural wrinkles, folds, drape, and subtle sheen under studio softbox light.
Style: professional photography studio fitting-test look, realistic natural human proportions, natural relaxed posture, softbox lighting, realistic skin texture, no AI feel.
Background: seamless photography studio gray, plain solid color with subtle natural gradient only, no lines, no patterns, no text, no labels, faint shadow only.
Negative constraints: no grid lines, no labels, no decorative background, no extra logos, no distorted hands, no extra fingers, no face changes between views, no clothing changes between views, no hairstyle changes, no plastic skin, no cartoon style.
Workflow
First Pass: Create the
- Read any explicitly referenced images or existing files.
- Look for relevant generated storyboard images in the current context and in the project directory.
- Decide whether the task is reference extraction or from-scratch design. If relevant storyboard imagery exists, prefer reference extraction.
- Choose or derive .
- Create the output directory if needed.
- Write using the template, including any storyboard reference paths.
- Include a complete image-generation prompt inside the .
- Ask the user to confirm or revise the spec before generating the image.
Second Pass: Generate the
Only after confirmation:
- Re-read the approved .
- Re-check the current context and project directory for relevant storyboard reference images, especially any paths listed in the approved spec.
- Generate the character design sheet image using the approved prompt with an image-generation model/tool. If relevant storyboard images exist, pass them as reference inputs and explicitly instruct the model to preserve the same character identity, wardrobe, props, and visual tone from those storyboard frames. Do not replace this with SVG, code drawing, canvas, HTML/CSS, or placeholder art.
- Save the image to .
- Inspect the image if possible.
- If the sheet fails critical requirements, regenerate or explain the issue.
- Final response should include both saved paths and render the image when supported.
Quality Checklist
Before asking for confirmation on the
, verify:
- The character name is filesystem-safe.
- The source mode is clear.
- Existing storyboard images in the current context and project directory were checked.
- Relevant storyboard reference paths are listed in the spec.
- Visual traits are specific enough to generate from.
- Uncertain extracted details are marked as assumptions.
- The sheet layout is specified as 16:9, two rows, 60/40 split.
- The prompt includes left/front/back/right views for both rows.
- The prompt describes clothing material details when clothing is visible.
- The prompt bans text, labels, grid lines, and decorative backgrounds.
Before finishing the
, verify:
- exists.
- Any relevant existing storyboard image was used as a generation reference.
- The image is 16:9.
- The sheet has two rows.
- Row 1 shows full-body left, front, back, and right views.
- Row 2 shows face/detail left, front, back, and right views.
- The background is plain photography-studio gray.
- Character identity, wardrobe, props, and hairstyle remain consistent.
- Clothing material details are visible and consistent where applicable.
- There is no visible text, label, grid line, logo, or unwanted pattern.
Example: Extract From Storyboard
User asks:
text
Create a character design sheet for the girl in storyboard/scene-01.png.
First create:
text
characters/campus-girl.md
The spec should describe:
- Teenage East Asian schoolgirl.
- Black low ponytail with loose bangs.
- White short-sleeve school sports polo.
- Blue athletic shorts.
- White socks with subtle blue accent.
- White running shoes.
- Fresh commercial campus-ad look.
- Studio gray 16:9 two-row turnaround sheet.
Then ask for confirmation.
After approval, generate:
text
characters/campus-girl.png