Tera Router

Image Generation

Text-to-image in OpenAI's images format — POST /v1/images/generations.

POST /v1/images/generations takes OpenAI's images request. The router resolves model like chat and rewrites only that field — size, n, quality, response_format, and any provider-specific extra reach the upstream untouched. The response is relayed unchanged.

curl http://localhost:8080/v1/images/generations \
  -H "Authorization: Bearer sk_tr_..." \
  -H "Content-Type: application/json" \
  -d '{ "model": "openai/gpt-image-1", "prompt": "A watercolor fox in a snowy forest", "size": "1024x1024" }'
{ "created": 1760000000, "data": [{ "b64_json": "iVBORw0KGgo…" }] }

Depending on the model and response_format, each item carries b64_json or a temporary url. With the OpenAI SDK, set baseURL to http://localhost:8080/v1 and call images.generate.

Providers

Every catalog provider that lists the image capability on the OpenAI dialect serves this endpoint — OpenAI, Gemini, Together AI, and Meta Muse — plus any OpenAI-compatible custom provider (for example a hosted FLUX endpoint addressed as <provider-slug>/<model>).

Slow models

Generation can take tens of seconds; the router waits up to 90 seconds before answering 504.

400 means the model names a provider without image generation; 503 means no account is connected for it.

On this page

Providers