ai-animation-production-skill

AI Animation Production Skill

A reusable, evolving production workflow for AI-assisted animation.

This repository is built around production reasoning, not one specific model or platform.

Scope Boundary

AI Animation Production Skill is currently a production orchestration and decision framework, not a fully autonomous animation generation engine. Future adapters and tools may connect an execution layer while SKILL.md remains the authoritative source for detailed Scope rules.

It helps answer questions such as:

• What must be locked before generation?
• When is a concept image ready to become a Master?
• When should one cinematic shot be split into several generation nodes?
• When should a new supporting still be created after video production has already begun?
• How should character entrances, camera direction, and dialogue ownership stay continuous?
• When should a result be trimmed, repaired, split, or fully regenerated?
• How can generation credits and time be protected?
• How should project state survive across sessions, tools, and agents?

Core Idea

A good AI-animation workflow is not a list of tools. It is a system for controlling story state, visual identity, continuity, cost, and failure.

The pipeline is iterative:

Project Setup
→ Story / Script Breakdown
→ Masters
→ Shot Planning
→ Still + Voice Assets
→ Video
→ Review
→ Repair / New Supporting Asset / Node Split
→ Editing
→ Export / Archive

The workflow may move backward when production reveals a missing state.

That is expected.

Important Rules

• One cinematic shot does not have to equal one generation node.
• The end of Node A should intentionally prepare the beginning of Node B.
• Do not rely on “continue the previous node” unless the platform actually persists state.
• A character should not appear from nowhere merely because the composition looks balanced.
• Supporting stills can be created just in time after video reveals a missing state.
• Speaker ownership and visible lip-sync requirements should be explicit.
• A good-looking draft is not automatically a Master.
• Use the default two-failure retry guardrail unless the Project Profile defines another policy.
• Minor defects should be solved in editing when that is safer and cheaper.
• Tool choice follows the production need, not brand loyalty.
• Research is not validation.

Visual Modes

The core workflow can support:

• 2D / Flat
• 3D
• Hybrid

Detailed full-2D and full-3D specializations are intentionally deferred until real production validates them.

Repository Structure

ai-animation-production-skill/
├── SKILL.md
├── README.md
├── CHANGELOG.md
├── STATUS.md
├── references/
│   ├── project-initialization.md
│   ├── story-script-breakdown.md
│   ├── character-workflow.md
│   ├── environment-workflow.md
│   ├── storyboard-shot-planning.md
│   ├── camera-continuity.md
│   ├── still-image-production.md
│   ├── voice-production.md
│   ├── video-production.md
│   ├── prompt-rules.md
│   ├── review-recovery.md
│   ├── editing-export.md
│   ├── file-naming-archive.md
│   └── tool-registry.md
├── templates/
│   ├── project-profile.md
│   ├── project-state.yaml
│   ├── shot-spec.md
│   ├── decisions.yaml
│   ├── handoff.md
│   ├── experience-update.md
│   ├── tool-record.md
│   └── failure-log.md
└── examples/
    └── example-project-profile.md

Quick Start

1. Read SKILL.md.
2. Determine the Complexity Level: LIGHT, STANDARD, or LONG-FORM.
3. Choose the minimum production structure justified by that level. A LIGHT task may need only an objective, necessary references, and a minimal Shot Spec; STANDARD and LONG-FORM projects should use the corresponding records defined in SKILL.md.
4. Define the necessary Visual Mode, source-of-truth assets, retry policy, and constraints.
5. Use references/story-script-breakdown.md and references/storyboard-shot-planning.md to turn story beats into shots and generation nodes when the project requires them.
6. Build only the Masters and assets needed for the first viable production pass.
7. Use the routing table in SKILL.md to open only the relevant reference module.
8. During production, record meaningful failures with templates/failure-log.md when the selected production structure requires it.
9. Promote only validated, reusable lessons into the Skill.

Tool Evidence

Tools are labeled by evidence level:

VALIDATED, EVALUATED, RESEARCHED, EXPERIMENTAL, DEPRECATED.

See:

references/tool-registry.md

The originating workflow validated several tools in real production and researched others as alternatives.

A tool being listed does not mean every user should adopt it.

Rights, Consent, and Provenance

Before publication or collaboration, confirm relevant rights for:

• voice cloning
• reference images
• footage
• music
• sound effects
• third-party character / design material

Record provenance when it matters.

Current Stage

This repository is a v0.2.0-pre workflow built from real AI-animation production experience.

The core production loop is complete enough for review and real project testing.

Intentionally deferred:

• dedicated LibTV CLI adapter
• LibTV internal Agent adapter
• full 2D specialization
• full 3D specialization
• additional platform adapters
• more case studies

Contributing / Iteration

When adding a rule, prefer:

Problem
→ Context
→ Cause
→ Tested Solution
→ Reuse Condition
→ Scope

Do not turn one isolated accident into a universal rule.

License

This project uses the Apache License 2.0.

Reuse and contribution rights are governed by the Apache License 2.0.

Third-Party Tools

Third-party product and project names are included only to document interoperability and production options.

Their inclusion does not imply affiliation or endorsement.

Capabilities may change; verify official documentation before following tool-specific setup instructions.