6-layer image prompts: who, action, where, camera, light, don'ts

Last updated 13 Oct 2026

The first frame decides the whole clip. Write prompts in 6 layers so the AI doesn't guess.

Course · Stage 5: Consistent AI images & video · Lesson 18/40 · 11 min

Why it matters

A short prompt like "a girl drinking tea" lets the AI pick everything — face, outfit, angle, light — so every scene looks different and nothing connects.

Still images are free or far cheaper than clips; fixing at the image step saves the most.

Step by step

1. Layer 1 — Who: a fixed character (age, ethnicity/nationality matching the market, hair, outfit, one unique mark).
2. Layer 2 — Action: one specific action in progress ("dropping ice into a cup"), not a list of things.
3. Layer 3 — Where: setting + key props, identical across scenes in the same place.
4. Layer 4 — Camera: shot size (close / medium / wide), angle, lens, vertical 9:16 or wide 16:9.
5. Layer 5 — Light & look: light source, time of day, style (phone photo / camera / cinematic).
6. Layer 6 — Don'ts: "no text, no logo, no watermark, exactly two hands per person, no extra fingers".

Common mistakes

• Changing the character description between scenes.
• Cramming three actions into one image.
• Skipping the don'ts, so images get random text or extra hands.

Ready-made prompt (ChatGPT / Claude / Gemini)

Write an English image prompt in 6 layers (who, action, where, camera, light, don'ts) for this scene: [scene]. Fixed character: [description]. Market: [US / UK / Vietnam…]. Frame [9:16 / 16:9]. End with: no text, no logo, no watermark, exactly two hands per person, no extra fingers.

Exercise

Write 6-layer prompts for 3 consecutive scenes with the same character, generate them and place them side by side.

Done when: Same person, same outfit, same setting in all 3; no stray text, no extra hands.

Need help? ngocdiep3135@gmail.com