GPT Image 2 character consistency: Picture books + checklist

Series illustration fails when characters morph. Anchor looks with GPT Image 2 plus a closing checklist from this series.

Series：(I) Model selection & parameters · (II) Prompt basics · (III) Advanced prompt techniques · (IV) Generation use cases · (V) Editing use cases

6.4 Picture books: keep the hero identical

Step 1 — Character anchor

Mint a canonical “casting sheet” still: lock

• Silhouette, wardrobe, palette
• Temperament / energy

Example: emerald hood ranger, walnut boots, satchel, kind guardian of woodland critters.

Every later spread references this sheet as law.

Step 2 — Story beats without drift

Keep feeding the same reference image. Reiterate in text:

• Same costume geometry + facial reads
• Same proportions + personality
• Only shift scene, blocking, plot motion

Sample line: “Snowy forest—ranger nurses an injured squirrel” while hood + boots stay locked.

Final checklist

Before every render/edit:

• Named the deliverable category (infographic / ad / portrait / book spread)?
• Documented scene · subject · hero details?
• Separated mutable vs immutable zones?
• Quoted on-image text + placement hints?
• Picked the right quality tier for this pass?
• Explained Image 1 / 2 / 3 when multi-plate?
• Only one creative lever this iteration?
• If it drifted, did I re-broadcast invariants?

Four principles (whole series)

1. Scaffold beats stacking adjectives
2. Explicit rails beat vibes
3. Short loops beat mega-prompts
4. Repeat the untouchables

Elite prompts answer do / don’t / must-stay simultaneously.