You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[bug] models: chatgpt-image-latest is not treated as gpt-image, so image tool requests are rejected #219
When chatgpt-image-latest (a GPT Image model listed by the account's /v1/models) is configured under the "OpenAI compatible" service, it does not get the openai-gpt-image profile added in #155 and falls back to the generic compatible profile. As a result, image tools such as smart edit, outpaint and object removal still send response_format=b64_json, and OpenAI rejects every request with HTTP 400. The model also loses the gpt-image capabilities (for example up to 16 reference images). Canvas image generation nodes do not send this parameter and are not affected.
Steps to reproduce
Settings → Model connections → Image → OpenAI compatible, Base URL https://api.openai.com/v1, Model ID chatgpt-image-latest.
Run "Smart edit" on any image node.
Expected vs. actual
Expected: treated like gpt-image-* (openai-gpt-image profile), requests omit response_format, and the tool returns an image.
Actual: the profile is compatible, tool requests include response_format, and OpenAI answers 400 unknown_parameter: Unknown parameter: 'response_format'.
Root cause and evidence
Verified on the latest origin/main (520ce9ed). Reproduced at the adapter level and with real API requests; the full UI flow was not exercised.
input_rules only matches name.startswith('gpt-image'). Running it gives profile=compatible, maxImages=8 for chatgpt-image-latest, versus profile=openai-gpt-image, maxImages=16 for gpt-image-2.
Real OpenAI API (2026-11-08), model chatgpt-image-latest:
/v1/images/generations with response_format → HTTP 400 unknown_parameter; the same request without it → HTTP 200.
/v1/images/edits with the exact form the tool sends (image field plus response_format) → HTTP 400 unknown_parameter.
Acceptance criteria
chatgpt-image-latest (and other chatgpt-image-* models) get the openai-gpt-image profile, and neither image tools nor generation requests send response_format.
Requests for gpt-image-* and for other OpenAI-compatible image services (Z-Image, Qwen-Image, etc.) are unchanged.
changed the title [-][bug] 模型接入:chatgpt-image-latest 未识别为 gpt-image,图片工具请求被拒绝[/-][+][bug] models: chatgpt-image-latest is not treated as gpt-image, so image tool requests are rejected[/+]on Nov 7, 2026
Activity
[-][bug] 模型接入:chatgpt-image-latest 未识别为 gpt-image,图片工具请求被拒绝[/-][+][bug] models: chatgpt-image-latest is not treated as gpt-image, so image tool requests are rejected[/+][bugfix] providers: treat chatgpt-image models as gpt-image (ThinkFlo…