Arena.ai & Gemini web (IMage)

[umershabbir40]

Hello love omniroute Can I please get info if there is already or are you planning to implement image generation using gemini web as the provider.

and can we fix arena.ai it is not working at all.

[diegosouzapw]

Hi @umershabbir40, thanks! Two separate answers.

Image generation with gemini-web

It existed, and it was deliberately removed. #10494 added an image path for gemini-web in August, and #11708 retired it a few weeks later: the parser was on a provenance and licensing hold, so the capability was pulled until it can be cleared or reimplemented independently. Chat, cookies, session rotation and the three text models stayed.

So today the connection carries text only, and /v1/images/generations has nothing to dispatch to it. There is no date for bringing it back — it depends on that review, not on the routing work.

Image generation itself is well supported, around 45 providers in open-sse/config/imageRegistry.ts. For the Gemini image family the working routes are:

• Antigravity: gemini-3.1-flash-image.
• OpenRouter: google/gemini-3.1-flash-image-preview and google/gemini-3-pro-image-preview.
• NanoBanana, fal.ai and UC, which carry the same family under their own ids.

Call them through POST /v1/images/generations.

Arena

It is still in the registry as lmarena, alias lma, using an arena.ai web session with a cookie. It was not removed. The known breakages — redirects, cookie handling, a decoder crash — are all fixed and closed, and there is no open issue about it right now, so "not working at all" is something I cannot reproduce from here.

Please open an issue with:

1. Your OmniRoute version.
2. Whether it is chat or image. Image on lmarena has no dedicated handler and falls through the generic OpenAI path, so it likely does not work.
3. The exact error from Monitoring → Logs, plus the HTTP status.
4. How you obtained the cookie. The one that matters is arena-auth-prod-v1.

With that it becomes a fixable report.

[umershabbir40]

For image generation you mentioned: Antigravity: gemini-3.1-flash-image.
But I can't see this and also tried making a call to it with suggested endpoint did not work. I have google Ai PRo plan with antigravity all other chat models are working and I dont see this model in provider quota page aswell I see other model names but not this please exlpain is it an issue on my end or this model is not supported any longer

[umershabbir40]

For image generation you mentioned: Antigravity: gemini-3.1-flash-image.
But I can't see this and also tried making a call to it with suggested endpoint did not work. I have google Ai PRo plan with antigravity all other chat models are working and I dont see this model in provider quota page aswell I see other model names but not this please exlpain is it an issue on my end or this model is not supported any longer

[diegosouzapw]

It is on the OmniRoute side, and it is a listing problem more than a model problem.

gemini-3.1-flash-image is registered as an image model, not a chat model (open-sse/config/imageRegistry.ts:379-386). Two consequences:

• The quota page tracks chat models, so an image model never shows up there. Nothing wrong on your end.
• In the catalog it appears with type: "image" and the full id antigravity/gemini-3.1-flash-image, and only when your Antigravity connection is eligible for it.

Check what your instance actually exposes:

curl -s http://localhost:20128/v1/models | jq '.data[] | select(.type=="image") | .id'

If antigravity/gemini-3.1-flash-image is in that list, call it on the image endpoint, not on chat completions:

curl -X POST http://localhost:20128/v1/images/generations \
  -H 'content-type: application/json' \
  -d '{"model":"antigravity/gemini-3.1-flash-image","prompt":"a red cube on a white table","n":1,"size":"1024x1024"}'

Reply with two things and I can tell you which of the two cases you are in: whether that model appears in the first command, and the full error body from the second. "Did not work" can be an empty catalog entry or an upstream rejection, and those have different fixes.

[diegosouzapw]

@umershabbir40 -- still happening on the latest version (3.8.52)? Otherwise I'll close this in 7 days for inactivity.