summary
OpenClaw appears to support gpt-image-2 in the shared image_generate capability now, but only through the OpenAI API-key path.
Today the bundled OpenAI image provider defaults to gpt-image-2, while provider inventory reports that openai/* image generation requires OPENAI_API_KEY. In contrast, OpenClaw already supports ChatGPT/Codex OAuth for openai-codex/* text models.
It would be very useful to support gpt-image-2 via OpenAI Codex OAuth as well, so users signed in with Codex/ChatGPT can use image generation without separately configuring OPENAI_API_KEY.
current gap
image_generate lists an OpenAI provider with default model gpt-image-2- that provider requires
OPENAI_API_KEY
- Codex OAuth works for
openai-codex/* model access, but not for image generation
- result: users can see that
gpt-image-2 is supported, but cannot use it through the same OAuth path they already use for Codex text models
requested behavior
Please add support for gpt-image-2 image generation through Codex OAuth.
A clean shape might be either:
- a separate
openai-codex image-generation provider/backend, or
- extending the existing OpenAI image provider to route through the Codex/Responses image generation path when Codex OAuth is available
Either way, it would be ideal to preserve the current API-key-based OpenAI provider while also allowing OAuth-based image generation.
reference implementation
Hermes just shipped a version of this:
Relevant details from that PR:
- adds a separate
openai-codex image backend
- exposes the same
gpt-image-2 catalog as the API-key-based OpenAI backend
- routes generation through the ChatGPT/Codex Responses
image_generation tool path
- works whenever the user is signed in with Codex OAuth
- does not require
OPENAI_API_KEY
- keeps the existing API-key OpenAI backend unchanged
That seems like a very good precedent for OpenClaw.
why this matters
- better UX: one auth story for Codex users
- avoids requiring both Codex OAuth and an OpenAI API key
- matches user expectations when
gpt-image-2 is already surfaced as a supported image model
- brings OpenClaw closer to parity with other Codex-integrated agents
Thanks — happy to test if this lands.
summary
OpenClaw appears to support
gpt-image-2in the sharedimage_generatecapability now, but only through the OpenAI API-key path.Today the bundled OpenAI image provider defaults to
gpt-image-2, while provider inventory reports thatopenai/*image generation requiresOPENAI_API_KEY. In contrast, OpenClaw already supports ChatGPT/Codex OAuth foropenai-codex/*text models.It would be very useful to support
gpt-image-2via OpenAI Codex OAuth as well, so users signed in with Codex/ChatGPT can use image generation without separately configuringOPENAI_API_KEY.current gap
image_generatelists an OpenAI provider with default modelgpt-image-2OPENAI_API_KEYopenai-codex/*model access, but not for image generationgpt-image-2is supported, but cannot use it through the same OAuth path they already use for Codex text modelsrequested behavior
Please add support for
gpt-image-2image generation through Codex OAuth.A clean shape might be either:
openai-codeximage-generation provider/backend, orEither way, it would be ideal to preserve the current API-key-based OpenAI provider while also allowing OAuth-based image generation.
reference implementation
Hermes just shipped a version of this:
Relevant details from that PR:
openai-codeximage backendgpt-image-2catalog as the API-key-based OpenAI backendimage_generationtool pathOPENAI_API_KEYThat seems like a very good precedent for OpenClaw.
why this matters
gpt-image-2is already surfaced as a supported image modelThanks — happy to test if this lands.