# AI Image Generation Prompting — a skill for stills that hold video continuity
Source: ekly.ai/guides/ai-image-prompting — free to use with any AI assistant.

## Hard rules (pillar)
1. Lock faces, wardrobes, rooms, and props as stills before generating video.
2. Universal formula: Subject identity + Pose/crop + Environment + Lighting + Lens/framing + Constraints.
3. Sweet spot ~40–120 words; put identity locks and lighting in the first half.
4. Choose the model by job (character sheet, location plate, multi-ref continuity, identity edit, mask inpaint, on-image text) — not by brand loyalty.
5. Generate stills at the video's aspect ratio when they'll become first-frame references.
6. If reference conditioning silently drops mid-session with the same inputs, switch models — don't reword a third time.
7. For marketing/web heroes: direct **creation happening**, rich saturated color, a platform-specific cue, and premium cinematic light — not a lone object on a dead background.
8. Light/dark pairs must keep the same subject, composition, and energy; only background/light changes.

## Family chapters
- Gemini Nano Banana — ekly.ai/guides/ai-image-prompting#nano-banana-and-gemini
- Seedream / BytePlus — …#seedream-and-byteplus
- GPT Image — …#gpt-image
- FLUX — …#flux
- Ideogram & Recraft — …#ideogram-and-recraft
- Marketing & web creative — …#marketing-and-web-creative

## Related
- Continuity craft — ekly.ai/guides/ai-video-continuity
- GPT Image 2 in Ekly — ekly.ai/models/gpt-image-2