Character Design Runtime

This is the lean runtime entry for character design. The full manual remains available as on-demand knowledge, but the default path should produce a usable character packet fast.

Core principle: make the character memorable, stable, and filmable. Do not optimize for prettiness; optimize for difference, silhouette, behavior, identity permission, and repeatable visual anchors. For production-critical roles, treat the character card as a frontloaded asset: the reference image is already part of the prompt. The workflow is model-agnostic; adapt the prompt wording to the user's image or video tool.

When an existing character reference or unusual body shape matters, do not start with a finished character sheet. First lock the base asset: proportion / height, face, permanent props, then W0 turnaround. A full sheet with expressions, wardrobe, skills, and text comes only after those bases pass QC.

For multi-shot or multi-segment AI video, separate immutable identity anchors from mutable character state. The same character should keep silhouette, face/hair, costume category, signature prop, and posture habit, while body state, emotion, fatigue, damage, prop ownership, gaze, and action residue evolve visibly across segments.

For cross-scene or cross-style video work, make a character keyframe or character sheet before video prompting. The sheet should lock identity anchors, expression range, costume, permanent wearable props, and the hand / body surfaces that interact with those props. The video model should animate a solved asset, not invent the role design inside the action prompt.

For audience-facing protagonists, add a likability pass before polishing design. A character becomes easier to follow when the first scene visibly gives at least one of: unfair treatment, kindness toward a specific subject, or distinctive ability / success potential. These must become actions, object relationships, posture, props, or skills, not personality labels.

For specialized live-action or short-drama asset libraries, treat each image as casting + wardrobe fitting + identity proof. Read the short-drama references only when the user explicitly asks for casting libraries, vertical drama roles, or episode-ready short-drama production.

Output Contract

For most tasks, produce a compact character packet: