Allow compatible providers to opt into standalone image generation

Resolved 💬 2 comments Opened Aug 7, 2026 by linshuboy Closed Aug 7, 2026
💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

What variant of Codex are you using?

CLI

What feature would you like to see?

Allow model providers that implement the OpenAI-compatible Images API to opt into Codex's standalone image_gen.imagegen tool.

An explicit provider capability could enable the feature:

[model_providers.example]
image_generation_mode = "standalone"

The capability should remain disabled by default for providers that do not declare it. When enabled, standalone image generation and editing would use the active provider's configured OpenAI-compatible base URL and authentication, calling the standard images/generations and images/edits paths.

Additional information

Codex already has the standalone image-generation extension and the Images API request types. The remaining limitation is that tool exposure is tied to the built-in provider/authentication gate, so a provider that implements the same compatible Images API cannot explicitly declare support.

The proposed change keeps the existing ChatGPT backend authentication checks for the built-in OpenAI provider, adds an opt-in capability for compatible providers, carries it through remote thread configuration, and defaults to disabled for backward safety.

A draft implementation with focused tests is available here:

https://github.com/linshuboy/codex/pull/1

Would this direction align with the Codex roadmap? If so, please invite an upstream PR from the linked branch.

View original on GitHub ↗

2 Comments

github-actions[bot] contributor · 21 days ago

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

  • #36832

Powered by Codex Action

linshuboy · 21 days ago

Thanks for flagging the overlap. This request is covered by the earlier canonical proposal in #24465, and #36832 tracks the same CLI symptom.

I am closing this issue in favor of #24465. For reference, a focused draft implementation is available in the contributor fork:

https://github.com/linshuboy/codex/pull/1

The implementation uses an explicit opt-in for providers that implement the OpenAI-compatible Images API, keeps the default disabled, and preserves the existing built-in provider authentication checks.