Communitygithub.com

3d-animation-short-generator

Create complete stylized 3D animated shorts (Pixar-style) from a story idea through an ordered production workflow: project brief, story outline, character and environment cards (Nano Banana Pro), standardized six-column shot table, text storyboards, Veo 3.1 single-shot clips, ffmpeg assembly, BGM, and final review. Use when the user wants an end-to-end narrative animation with strong character consistency, scene continuity, timing, camera, performance, and audio control. Not for single images, simple edits, photorealistic live action, or one standalone clip.

What is 3d-animation-short-generator?

3d-animation-short-generator is a Claude Code agent skill that create complete stylized 3D animated shorts (Pixar-style) from a story idea through an ordered production workflow: project brief, story outline, character and environment cards (Nano Banana Pro), standardized six-column shot table, text storyboards, Veo 3.1 single-shot clips, ffmpeg assembly, BGM, and final review. Use when the user wants an end-to-end narrative animation with strong character consistency, scene continuity, timing, camera, performance, and audio control. Not for single images, simple edits, photorealistic live action, or one standalone clip.

Works with✓Claude Code~Codex CLI~Cursor✓Antigravity✓Gemini CLI
npx skills add https://github.com/rizkifahtur/antigravity-3d-animation-skill/tree/HEAD/.agents/skills/3d-animation-short-generator

Ask in your favorite AI

Open a new chat with this agent skill pre-loaded.

Documentation

3D Animation Short Generator — Antigravity Edition

Port of MiniMax Hub's 3d-animation-short-generator (v0.5.4) for Google Antigravity. The story/shot/storyboard methodology is unchanged; the platform layer is replaced:

MiniMax Hub originalAntigravity edition
Canvas nodesFiles in animation/<slug>/ inside the workspace
Choice cardsNumbered option list in chat → STOP and wait for the user's reply
hub_generate_imageAntigravity built-in image generation (Nano Banana Pro). Fallback: scripts/gen_image.py
hub_generate_video with H3 / Seedance 2.0scripts/veo_generate.py → Veo 3.1 (final) / Veo 3.1 Fast (draft) via Gemini API
Hub assembly + BGMscripts/assemble.sh (ffmpeg, BGM ducking, aspect-ratio crop)

Talk to the user in the language they write in (e.g. Bahasa Indonesia). Keep file names and prompts to video/image models in English.

Core rule: story first → choice prompt for aspect ratio + total duration right after intake → ordered project files → every required confirmation via a choice prompt → after character/scene cards, fixed order: six-column shot table with per-second directives + audio cues + spatial anchor chain → Step 5.5 self-check → one text storyboards document (one section per shot) → video-model + resolution choice → keyframe + Veo clip per shot → assembly with BGM. The final video must contain no storyboard artifacts.

Antigravity Setup (check once per project)

  1. Agent model (Antigravity model dropdown): use the strongest reasoning model available — Gemini Pro tier (High thinking) or Claude Opus/Sonnet — for Steps 1–6 (story, shot table, storyboards). A Flash-tier model is fine for Step 7–8 file/script work. See references/model-selection.md.
  2. Mode: use Planning mode (not Fast) so the agent writes plans/artifacts and pauses for review.
  3. Video/API prerequisites (needed from Step 7): ask the user to confirm, or check in the terminal:
    • GEMINI_API_KEY exported in the shell (from Google AI Studio, billing enabled — Veo is paid).
    • pip install google-genai and ffmpeg on PATH.
    • Never print, log, or write the API key into project files.
  4. If prerequisites are missing, Steps 0–6 can still run; stop before Step 7 and tell the user what to install.

Choice Prompt Format (replaces choice cards)

Whenever this skill says "choice prompt", output exactly this shape and then end your turn:

**[Gate name]** — pilih satu (balas angka):
1. Option A (recommended)
2. Option B
3. Option C
0. Other / custom — tulis sendiri

Never auto-select, never continue past a gate in the same turn, never replace a gate with prose. The user may answer with a number or free text.

Project Folder Layout (replaces canvas)

Create animation/<slug>/ in the workspace (slug = kebab-case working title). Write every durable artifact to disk in this order — do not dump long production content only in chat:

animation/<slug>/
  00-project.json            # approved settings: aspect, duration, resolution, video_model, style, language
  01-brief.md                # Project Brief
  02-story-outline.md        # Story Outline
  03-characters/             # <name>-card.png + <name>-card.md (visual-ID note + prompt used)
  04-scenes/                 # <scene>-card.png + <scene>-card.md (landmarks, light states, prompt)
  05-shot-table.md           # six-column table; first line "shot-table self-check: passed" once passed
  06-storyboards/
     text-storyboards.md     # one "## S01" section per shot (authoritative)
     S05-extracted.md        # only for shots flagged for heavy iteration
     pencil/S01.png          # only in visualization mode
  07-keyframes/S01-start.png # first frames generated from cards (keyframe mode)
  08-clips/S01_v1.mp4        # versioned; highest version = current
  09-assembly/main.mp4, bgm.*, final.mp4
  10-review.md

When regenerating, bump the version suffix (_v2) — never overwrite — and state which version is current. 00-project.json is the single source of truth for settings; update it after each gate.

Global Visual Style Lock

Unless the user explicitly requests another visual style, all character cards, scene cards, shot tables, storyboards, keyframes, clips, and final composites use this style:

  • Rendering: Pixar-inspired 3D cartoon rendering, C4D + Octane renderer look, high-end animated feature quality.
  • Character design: exaggerated geometric simplification balanced with excellent material detail; large readable silhouettes; Q-version proportions when appropriate (2.5–3 heads tall for cute/childlike characters).
  • Hair/fur: sculpted clumps and clean block shapes plus fine flyaway edge hairs / fuzzy rim detail.
  • Skin/material: warm subsurface scattering, soft translucent red through ears, cheeks, nose, fingertips; no hard plastic skin.
  • Acting: exaggerated, lively Disney/Pixar performance — squash and stretch, strong brows, eye corners, pupils, lips, cheeks.
  • Motion: high-energy poses, clear line of action, anticipation, fast-but-readable timing, elastic body mechanics, vivid micro-expressions.

Negative style: no photorealistic live action, no flat 2D anime, no plastic toy skin, no stiff mannequin posing, no lifeless faces.

STEP 0: Intake and Project Plan

Capture: one-line idea; desired output depth (blueprint only / assets only / shot table / storyboards / clips / assembled / final with BGM); approximate length; aspect ratio; tone (default warm stylized 3D); dialogue mode (dialogue / voiceover / no speech); dialogue language only if the user states it — do not default to English.

Immediately after intake, before anything else, show two choice prompts in one message (user answers both):

Aspect ratio:

  1. 16:9 landscape (recommended — native Veo)
  2. 9:16 vertical short (native Veo)
  3. 1:1 square (rendered 16:9, center-cropped in assembly)
  4. 4:5 social portrait (rendered 9:16, cropped in assembly)
  5. Custom

Total duration:

  1. 30–60 s (recommended)
  2. 15–30 s
  3. 60–90 s
  4. 90–180 s (many clips → higher Veo cost; warn the user)
  5. Custom

Proceed only after both are answered. Save to 00-project.json and reuse everywhere.

STEP 1: Project Brief → 01-brief.md

Include: working title, one-line What-if, emotional premise, target audience feeling, deliverables, approved aspect ratio, approved duration, estimated shot count (duration ÷ ~6 s), dialogue mode and language (language not specified if not given), initial risks, dialogue intent.

Choice prompt: 1. Continue with this direction (recommended) · 2. Regenerate premise options · 3. Revise emotional premise · 4. Refine dialogue direction.

STEP 2: Story Outline and Gates → 02-story-outline.md

Include: protagonist Want / Need / flaw; core world rule; 8-beat causal story spine; emotional anchor and payoff; dialogue beats if requested; red-line checks.

Gate checks (write pass/fail per item in the file): protagonist is active · crisis intensified by protagonist's flaw · coincidence never solves the problem · ending reuses an earlier emotional anchor · antagonistic pressure is not a flat villain · dialogue reveals relationship change instead of explaining theme.

Choice prompt: 1. Approve story and continue (recommended) · 2. Revise beats · 3. Revise emotion curve · 4. Revise dialogue beats · 5. Return to premise.

STEP 3: Character Cards → 03-characters/

Generate with Antigravity's image generation (Nano Banana Pro). Order: protagonist → contrast/pressure character → optional supporting. Each card is a 16:9 production reference sheet with readable labels:

  • Name label + role label; main 3/4 view; front / side / back views; expression row; costume/material/prop details with prop labels.
  • Write <name>-card.md next to the PNG: age range, body type, hairstyle, outfit colors, signature props, do-not-change traits, and the exact prompt used (identity lock, reused verbatim later).

If built-in image generation is unavailable or saves elsewhere, run python3 scripts/gen_image.py (see script header) or copy the output into this folder.

Choice prompt: 1. Lock character designs and continue (recommended) · 2. Regenerate protagonist card · 3. Adjust specific visual details · 4. Add another character card. Warn: changing locked designs later forces regeneration of the shot table onward.

STEP 4: Scene Cards → 04-scenes/

Environment only — no characters, people, silhouettes, hands, or faces. Include: main environment overview, key light states (day/night), emotional sub-spaces, continuity landmarks (fixed objects whose screen position must persist, e.g. door frame, sofa, tree), important props. Document landmarks and lighting in <scene>-card.md.

Choice prompt: 1. Lock scene design and continue (recommended) · 2. Regenerate scene card · 3. Add another scene angle · 4. Adjust lighting or layout.

STEP 5: Standardized Shot Table (Six Columns) → 05-shot-table.md

Mandatory; cannot be swapped with storyboard or video steps. Read and follow references/shot-table-spec.md for the exact schema, per-second directives, table rules, approval prompt, and the Step 5.5 self-check.

Antigravity/Veo constraints that override the original spec:

  • Every shot duration is 4, 6, or 8 seconds (Veo 3.1 clip lengths). Longer beats are split into multiple rows.
  • In Shot Description, add render_mode: keyframe | reference (default keyframe) and optional video_model: veo-3.1 | veo-3.1-fast.
  • Dialogue lines must fit inside one clip; Veo generates audio per clip, so do not let a sentence span two shots.

Run the Step 5.5 self-check. Do not enter Step 6 until it passes.

STEP 6: Text Storyboards (default) + Pencil Storyboards (opt-in) → 06-storyboards/

Read and follow references/storyboard-guidelines.md. Show the storyboard-mode choice prompt first. Default = one authoritative text-storyboards.md with one section per shot. Pencil images are opt-in, human-review only, and never override the text. Extract a shot to S0X-extracted.md only when the user flags it for heavy iteration.

STEP 7: Video Model + Resolution Choice → Keyframes → Veo Clips

Read references/model-selection.md (models, modes, prompt shaping, cost) and references/fallback-policy.md (retry ladders, drift).

Minimum contract:

  • Show the video-model choice prompt, then the resolution choice prompt. Store in 00-project.json.
  • Before rendering everything, render S01 only as a test, show it, and ask approval (cost control).
  • Default render mode is keyframe: generate 07-keyframes/SXX-start.png with Nano Banana Pro from the exact character card(s) + scene card + the shot's first-second directive, then call scripts/veo_generate.py --image with it.
  • Build every clip prompt from the matching storyboard section (or extracted file), the identity lock text from the character .md, and the landmarks/lighting from the scene .md. Strip all storyboard labels ([char:…], [scene:…], [shot:…], [dur:…], [hook:…]), shot numbers, arrows, and notes.
  • Clips must be clean full-color 3D animation: no sketch texture, no labels, no subtitles unless requested, no watermarks.
  • On drift from Reference Anchors, follow references/fallback-policy.md; never silently assemble wrong clips.

After all clips render, list them in shot order with version numbers and show the clip-approval choice prompt from references/model-selection.md.

STEP 8: Assembly, BGM, Final Output → 09-assembly/

Read references/qc-checklist.md for assembly rules, BGM rules, final review checks, gate discipline, and latest-asset discipline.

  • Assemble with scripts/assemble.sh (exact shot order from the approved table; latest approved version of each clip; crops to 1:1 / 4:5 if chosen).
  • BGM: Veo produces per-clip audio only, so one continuous BGM track is either (a) a file the user provides, or (b) a track the user makes in a music tool (e.g. Lyria / Google Flow / any royalty-free library) — ask with a choice prompt. The script ducks BGM under clip audio (dialogue/SFX).
  • No subtitles or text unless requested. Final video must contain no storyboard traces.
  • Run the final review checks, write 10-review.md, and deliver the path of final.mp4.

Boundaries

Do not use this skill for a single image, a simple edit, a single clip, logo design, or pure prompt consultation.

Related Skills