# 0010: Sync unified GPT Image controls and remove moderation

- Status: Accepted
- Date: 2026-09-24
- Owners: Maintainers

## Context

The standalone GPT Image node supports two additional GPT Image 2.5 quality levels and counts 1–8. The unified panel used four quality levels, counts 1/2/4/9, and exposed moderation after the background control. Removing a middle widget would otherwise shift saved batch and save settings.

## Decision

`O1keyImageGenerator` uses the standalone GPT Image quality resolver for GPT Image 2 and 2.5. The panel offers counts 1–8 for GPT models and retains the older 9-image value when loading a saved workflow. New GPT panel selections default to PNG and smart resize; existing saved output and resize values remain intact. Other model families keep their count and default behavior.

The unified generator no longer exposes or submits `内容审查强度` / `moderation`. Its old positional widget at index 14 is removed after the existing old-layout migrations finish, preserving batch and save widget values at their new positions. Legacy panel job payloads that contain `moderation` are ignored. The provider client retains optional low-level support for callers outside the unified node.

## Compatibility

The node ID, earlier widget positions, model-to-route mapping, and provider request scheduler stay stable. The workflow migration is idempotent for both old 24-value arrays and new 23-value arrays. GPT Image continues to send one provider request per selected output image with `n=1`.

## Validation

Offline node, job, standalone GPT, and frontend tests cover quality conversion, count options, omission of moderation, and the saved-workflow migration.