Lira Image Prompts

Always read references/core-workflow.md. Classify the task, then completely read the union of applicable references:

• Character/sheet/portrait, Soul ID, Soul 2.0, AI Cast: character-generation.md plus generation-style.md.
• Location/environment/still/establishing frame, Soul Cinema: scene-location-generation.md plus generation-style.
• Prop/product object/exact object text: prop-generation.md plus generation-style.
• Existing-frame edit, NBP, Seedream texture repair, GPT Image 2 surgery: image-editing.md.
• Existing-location reverse angle/new camera position: scene-location-generation plus image-editing.
• Comprehensive or mixed review: all applicable references; load all only for a true full-system task.

Do not load unrelated task references. Keep project building blocks consistent and preserve core routing, edit order, anti-fail rules, response format, and user-language behavior. Hand complete Seedance video-prompt assembly to a dedicated video-director skill when available.