Creative Claw — GPT Image 2

Use image/gpt-image-2 when precise instruction following, typography, transparency, or controlled editing matters more than the speed and cost advantages of the default image model.

Core workflow

1. Define the final use, canvas shape, hierarchy, exact visible copy, transparency requirement, and protected details.
2. Search or import durable Creative Claw references.
3. Choose generation or edit mode. Any image_url or extras.image_urls makes the request an edit/composition.
4. Call get_model_params({ model: "image/gpt-image-2" }) immediately before generation; the live route may expose different optional controls.
5. Put the principal source or canvas in image_url. Put additional ordered references in extras.image_urls.
6. Set agentic_prompting: false for exact copy, reference-number mappings, masks, identity locks, hex colors, or a finished production prompt.
7. Use low or medium quality while exploring and high for an approved direction. Request several images only when the user wants alternatives.
8. Inspect text, identity, reference roles, masked boundaries, transparency, anatomy, product geometry, and composition.

Current Creative Claw contract