character-generation
Character Generation
Write one image-generation prompt for a wide landscape character reference sheet: wardrobe (front + back) + face identity for downstream image and video work.
Output is a JSON image prompt only. Return one valid JSON object in a json code block and stop. Do not generate, render, preview, or offer to generate the image. A required smile variant is a second, separate JSON object in its own json code block.
Use the structured JSON style in references/prompt-blueprint.md. Emit parseable JSON with double-quoted keys and strings; no comments, markdown inside the object, trailing commas, or unresolved placeholders. Omit irrelevant fields instead of filling them with guesses. Keep final_generation_instruction consistent with the structured fields.
Animated mode
Use this mode when the project is stylized 3D animation (the animated-video-production skill, or a request for Pixar-, DreamWorks- or feature-animation-style characters, places or props). The layout rules below still apply: one landscape image, the same panels, butted edge to edge with no divider lines. What changes:
- Render the character as sculpted, slightly stylized feature-animation CG: appealing proportions, large expressive eyes with catch-lights, shaped hair or groomed fur, simplified skin with soft subsurface shading. Replace the photoreal skin rules (pores, oil, fine lines) and
studio photographwording with this. Keep the studio sheet neutral and evenly lit. - Animals and creatures use the same three panels: headless front body, back body, and a large 3/4 head portrait. Describe fur colour, pattern and markings concretely, since they are the identity.
- A group of near-identical characters (a litter of kittens, a squad) can be one lineup sheet of complete characters instead of three panels. Say so in the prompt, and in every scene prompt that uses it.
- Scale: state the character's size against a familiar reference ("a small young cat, back at an adult's shin"). Scene prompts reuse it.
- Product mascots (a real product given a face and limbs): don't write the product from a description, because logos and labels get redrawn wrong. Pass the client's real product photo as the reference and ask for an edit that adds only the face, arms and legs, keeping the label, logo, colours and shape exactly; quote the label text. When the user already likes a take, edit that exact file, one take at a time.
- Where the prompt goes: in an animated project, save the prompt to the asset's
prompts/folder. It may then be generated with theflowskill; the prompt-only rule above covers standalone requests.