plotloom-character-reference-sheet

name	plotloom-character-reference-sheet
description	Creates production-grade character reference sheet briefs/prompts for Plotloom short dramas: turnaround, color palette, expressions, micro-expressions, head details, wardrobe/accessory details, props, and hand gestures for AI-video consistency.

Plotloom Character Reference Sheet

When to Use

Use when the user asks for a character reference sheet / 角色设定图 / model sheet / turnaround sheet / AI-video character consistency asset, or when ordinary character grids are not enough for stable video production.

Use this before video generation when a named character needs stable identity, wardrobe, expression acting, props, or hand/gesture continuity.

Inputs

• series.md and characters.md, if a Plotloom series repo already exists.
• User-approved character concept: name, role, age range, genre, personality, visual archetype, costume state, signature props.
• Optional existing images: character grid, portrait, style reference, storyboard keyframes, costume reference.
• Target output character slug, e.g. aelia, cassian, selene.

Outputs

Create or update a production asset folder such as:

assets/cast/<slug>/reference-sheet/
  brief.md
  prompt-imagegen2.txt
  negative-notes.md
  reference-inputs.json
  candidates/
  selected.png

For human-readable cast records, also update characters.md with the final selected path:

assets/cast/<slug>/reference-sheet/selected.png

Why This Format

A normal 3x3 character grid is useful for early visual exploration, but it often fails for AI video continuity because it emphasizes attractive portraits rather than production constraints. A full character reference sheet should lock:

• identity and silhouette
• front / 3-4 / side / back body views
• color palette
• expression progression
• micro-expressions
• head / hair details
• wardrobe and accessory construction
• signature prop design
• hand gestures and object-use gestures

This makes the character easier to reproduce across shots, angles, emotions, costumes, and video prompts.

Required Sheet Sections

Every production sheet prompt should ask for these panels unless the user deliberately wants a lighter sheet:

1. Character Metadata Block

   • Name
   • Alias / title
   • Role
   • Age range
   • Personality keywords
   • Speech / posture / body-language notes
   • AI-video continuity invariants

2. Main Identity + Scale Sheet

   • full-body front view
   • full-body 3/4 front view
   • full-body side view
   • full-body back view
   • silhouette / height guide
   • body proportion and posture notes

3. Color Palette

   • skin
   • hair
   • primary costume color
   • secondary costume color
   • metal / jewelry / weapon color
   • shadow / accent color

4. Expression Progression

   • neutral
   • guarded smile
   • wounded / betrayed
   • anger restrained
   • shock
   • fear hidden under pride
   • determination
   • cold victory / comeback

5. Micro Expressions

   • eye tension
   • brow changes
   • mouth-corner shifts
   • side glance
   • suppressed tears
   • suspicious look
   • pre-revenge calm

6. Head Detail Sheet

   • front head
   • 3/4 head
   • side profile
   • back hair view
   • hair ornament / crown / scars / curse mark details

7. Wardrobe / Accessories Details

   • collar / neckline
   • sleeves / cuffs
   • belt / waist detail
   • shoes / boots
   • cape / train / skirt structure
   • jewelry / family crest / curse mark
   • fabric/material notes

8. Prop Sheet

   • signature prop front view
   • side/back view when relevant
   • active / glowing / broken state
   • how the character holds or wears it

9. Hand Gestures

   • relaxed hand
   • fist
   • pointing / accusing
   • signing contract
   • holding prop
   • touching pendant / crest / curse mark
   • reaching / refusing / gripping

Prompt Rules

• Ask for one single clean reference-sheet image, not separate illustrations.
• Ask for readable panel layout with clear section blocks.
• Use minimal readable English section headers only; avoid long tiny captions that become gibberish.
• Do not include social-media UI, subtitles, watermark, logos, app overlays, likes, stars, or random numbers.
• Do not make the sheet a beauty poster. It is a production document.
• Preserve one outfit per sheet. If a character has multiple story states, make separate sheets:
  • nameless-maid
  • contract-duchess
  • true-heiress
• For AI-video continuity, include do-not-change invariants explicitly: face shape, hair silhouette, costume palette, signature prop, body language, and key accessories.
• For female-oriented short dramas, prioritize expressive acting references: betrayal, restrained hurt, comeback confidence, romantic guardedness, power shift.

Recommended ImageGen2 / Codex Prompt Template

Use this as the base for prompt-imagegen2.txt:

Create ONE production-grade CHARACTER REFERENCE SHEET for an AI-video short-drama character.

Character: <NAME>
Alias / title: <ALIAS>
Role: <ROLE>
Age range: <AGE>
Genre / market: <GENRE>
Personality keywords: <KEYWORDS>
Core contradiction: <CONTRADICTION>

Visual identity:
<FACE, HAIR, BODY, COSTUME, PALETTE, ACCESSORIES, PROP>

The sheet must be a clean professional animation / game / film concept-art reference sheet, not a poster, not a fashion illustration, not a social media screenshot.

Required layout, all in one vertical or wide production sheet:
1. CHARACTER METADATA BLOCK: name, alias/title, role, age, personality keywords, visual invariants.
2. MAIN IDENTITY + SCALE SHEET: full-body front view, 3/4 front view, side view, back view, plus a small silhouette/height guide.
3. COLOR PALETTE: 6-8 clean swatches matching skin, hair, main costume, secondary costume, metal/jewelry, accent/shadow colors.
4. EXPRESSION PROGRESSION: 8 bust/head expressions: neutral, restrained hurt, shock, anger restrained, suspicious side glance, suppressed tears, determination, cold comeback confidence.
5. MICRO EXPRESSIONS: close crops of eyes/brows/mouth showing subtle acting changes.
6. HEAD DETAIL SHEET: front, 3/4, side profile, back hair view, and key ornament/mark details.
7. WARDROBE / ACCESSORIES DETAILS: collar, sleeves, belt/waist, shoes, jewelry/crest/curse mark, fabric and construction closeups.
8. PROP SHEET: signature prop in normal and active/broken/glowing state, with how it is held/worn.
9. HAND GESTURES: relaxed hand, fist, pointing/accusing, signing contract, holding prop, touching key accessory, reaching/refusing gesture.

Style: <STYLE>. Consistent face and costume across every panel. Clean readable panel grid, production reference sheet, high detail, coherent proportions, no extra characters.

Hard constraints:
- no watermark, no logo, no social media UI, no Chinese subtitles, no random numbers
- no long unreadable micro text except simple section labels
- no inconsistent outfits between panels
- no duplicate different characters
- no distorted hands or extra fingers
- no beauty poster composition

Return JSON with image_path and notes.

Plotloom Command Shape

When using the current Plotloom image path, generate via plotloom image generate so the asset stays in the repo-first flow:

plotloom --repo /Users/haha/plotloom_repo/<slug> image generate \
  --kind cast \
  --character <character-slug> \
  --filename reference-sheet/candidates/v001.png \
  --prompt-file /Users/haha/plotloom_repo/<slug>/assets/cast/<character-slug>/reference-sheet/prompt-imagegen2.txt \
  --image /optional/existing-character-grid-or-style-ref.png \
  --timeout 600

After visual review, copy or regenerate the accepted image to:

plotloom --repo /Users/haha/plotloom_repo/<slug> image generate \
  --kind cast \
  --character <character-slug> \
  --filename reference-sheet/selected.png \
  --prompt-file /Users/haha/plotloom_repo/<slug>/assets/cast/<character-slug>/reference-sheet/prompt-imagegen2.txt \
  --image /optional/existing-character-grid-or-style-ref.png \
  --timeout 600

Or copy the accepted candidate manually:

cp assets/cast/<character-slug>/reference-sheet/candidates/v001.png \
   assets/cast/<character-slug>/reference-sheet/selected.png

Do not merely write prompt-imagegen2.txt and stop when the user expects an image; this skill is an image-production skill.

Workflow

1. Locate the active Plotloom series repo. If no repo exists, hand off to plotloom-series-brainstorming first.
2. Read series.md, characters.md, and any existing assets/cast/<slug>/ images.
3. Decide whether this is a first-pass sheet or a state-specific sheet (nameless-maid, contract-duchess, true-heiress, etc.).
4. Write assets/cast/<slug>/reference-sheet/brief.md with stable identity, story function, visual invariants, palette, expressions, wardrobe, prop, and hand gestures.
5. Write prompt-imagegen2.txt using the template above.
6. Write reference-inputs.json listing all source images used and their purpose.
7. Generate candidates with plotloom image generate directly; do not stop at prompt files unless the user explicitly asks for prompt-only work.
8. Review candidates visually for:
   • same person across panels
   • complete front/side/back body views
   • readable expression progression
   • useful micro-expression details
   • wardrobe and prop consistency
   • no watermark/social UI/subtitle overlays
   • no unreadable text dominating the sheet
9. Copy the accepted candidate to selected.png and update characters.md with the path.

Quality Bar

A good character reference sheet must answer:

• Can a video model keep this character consistent across front, side, back, and motion?
• Can a prompt writer describe this character without inventing missing costume details?
• Can an animator or AI video generator infer how this character emotes?
• Are the signature props and hand gestures clear enough for repeated story use?
• Does the sheet reduce ambiguity compared with a 3x3 beauty grid?

Failure Modes

• If the output is a collage of pretty portraits only: rerun with stronger production reference sheet / turnaround / wardrobe details / hand gestures wording.
• If tiny text becomes gibberish: keep only section headers and store real metadata in brief.md.
• If outfits change across panels: split into separate outfit-state sheets or reduce visual complexity.
• If face identity drifts between panels: use a selected portrait or character grid as a reference image and explicitly say same face in every panel.
• If hands are poor: generate a separate hand/prop sheet rather than forcing everything into one image.

Next Skill Handoff

• Use plotloom-shot-prompts after selected reference sheets exist and the user wants video prompts.
• Use plotloom-video-adapter when prompts are ready for Dreamina/Seedance generation.
• Use plotloom-asset-selection to accept/reroll generated sheet candidates.