character-design-continuity

What it does

Guides an agent to preserve recognizability of a recurring character across AI-generated media by treating continuity as a production system. It emphasizes creating a compact, operational character bible with defined identity locks, continuity tolerances, and reference packs, plus a structured approach to prompt planning, multi-shot continuity, and iteration/repair workflows.

How it works

• Start with a continuity intent: define Continuity target, Delivery form, and Allowed variation, plus hard invariants and rights status.
• Build the character bible: specify identity lock (name, body/read, silhouette, anchors for face, body, color, signature items, and negative identity).
• Define Continuity tolerances: Locked, Flexible, and Story-changing levels for elements like hair, jacket, age, and facial features, with examples.
• Create model-sheet substitutes and reference strategy: choose between text, single image, multi-reference pack, image-to-image, personalization, or rigged assets; include expression/pose ranges, costume/hair/makeup/prop sheets, palette chips, and do-not-generate sheets.
• Establish a reference strategy: collect light to heavy references, aiming for minimal drift.
• Provide a prompt planning framework: structure prompts into Character anchor, Scene, Continuity, Style, and Negative/Avoid sections; keep anchor wording stable.
• Manage multi-shot/serial continuity with a continuity ledger per shot: track shot ID, story day/time, costume/hair/makeup/prop state, emotion/pose, camera angle, references, seeds, approved/rejected notes, and repair instructions.
• Apply Continuity tiers to scope: from Recognizable concept (Tier 1) up to Deterministic production asset (Tier 4).
• Outline design ranges for expression and pose to prevent drift; specify performance limits and action constraints.
• Detail costume/hair/makeup/props with explicit naming, stateful notes for story changes, and guidance on how fabric, gear, and marks move across frames.
• Address rights and identity constraints: avoid generating recognizable living persons without permission; provide conservative rules for minors, deceased individuals, and public figures; maintain consent and provenance records.
• Define an Iteration and repair workflow: generate exploration batch, select canonical take, extract and publish the bible, generate reference views, run continuity batches, classify drift, and repair using narrow tools; repair one variable at a time when feasible.
• List common failures and repairs to guide QA and corrections.

When to use it

Use when you need to create, preserve, revise, or QA a recurring character across AI-generated media and campaigns, especially for building character bibles, reference packs, consistent prompts, expression/pose ranges, and continuity QA across multi-shot outputs.

What it can touch

Tools declared: claude-code, codex, copilot, cursor. It instructs planning, reference management, prompt structuring, and iterative repair workflows to touch prompts, reference packs, and continuity logs; it also defines acceptable artifact states and QA checkpoints.

Caveats

• Emphasizes that perfect identity preservation from text prompts is not guaranteed due to drift in image/video systems.
• Recommends consent, rights, and provenance considerations for identity, especially for living or real-person likenesses.
• Contains guidance that is policy-oriented (consent, rights) and procedural (ledgering, repair loops) rather than guarantees of outputs.