Add an opt-in Image Preflight mode that lets ChatGPT check a planned image request with the image-generation system before rendering.
The preflight could return three things:
- Policy: Can the request be attempted as written? If not, identify the specific problem and the smallest necessary change.
- Interpretation: How is the image system likely to understand the composition, and are any instructions ambiguous or conflicting?
- Reliability: Which parts are likely to render well, and which elements create elevated risk of anatomy, pose, hand, overlap, material, lighting, or composition failures?
A simple result could be:
GREEN — likely to work
YELLOW — allowed but technically fragile
RED — cannot be generated as written
The feature should be advisory, not restrictive. If a request is allowed but difficult, the user should still be able to attempt the original version. It should not silently rewrite prompts or automatically simplify creative concepts.
For any suggested modification, the system should identify:
- the specific risk,
- the smallest proposed change,
- the likely artistic tradeoff.
Possible modes could include Policy Check, Reliability Check, and Full Preflight.
The goal is to reduce wasted generations and unnecessary prompt rewriting while preserving experimentation and user intent.