-
1
Start with an adult-only fictional brief that names the scene goal, subject age clarity, visual style, wardrobe boundary, camera framing, lighting, and reuse context.
-
2
Keep the prompt generator role narrow: it should organize a reviewable brief, not produce explicit prompt dumps, filter-bypass language, or copied prompt packs from X.
-
3
Route still-image planning through verified labels such as `wan2.6-image`, `wan2.7-image`, and `wan2.7-image-pro` when the source frame needs creation or repair.
-
4
Add `wan2.6-i2v` notes only after the still frame is reviewed and the motion handoff needs continuity, duration, camera movement, and preservation instructions.
-
5
Rewrite third-party Wan prompt or ComfyUI discussion into product-safe planning notes instead of copying exact wording, checkpoint names, media URLs, or download instructions.
-
6
Review the generated result for adult-only clarity, real-person likeness risk, protected-character similarity, moderation ambiguity, prompt leakage, and local reuse rights.
-
7
Stop the workflow when the request depends on private photos, celebrity resemblance, unclear age signals, coercive framing, external LoRA installation, or unsafe bypass instructions.