Configure providers and models
A provider decides where the request is sent. A model decides what the request can do. Start with one stable provider.
Image providers
| Provider | Typical use | Considerations |
|---|---|---|
| SiliconFlow / Kolors | General images and a stable image-to-image path | Uses a fixed set of sizes |
| ModelScope | Qwen Image, FLUX Krea and Z-Image variants | Parameter support differs by model; Gateway does not force one common size |
| OpenAI-compatible Image | OpenAI or a private compatible endpoint | Configure Base URL, key and model explicitly |
| Pollinations | Public or experimental image routing | Availability and size behavior can change |
| ByteDance Seedream | Optional image provider | Depends on account and model access |
| Agnes Image | Generation, reference images and editing | Select explicitly; it is not a universal default |
Video providers
The main current video route is Agnes Video v2.0 for text, image and keyframe inputs. A provider should not appear in the UI until it has a real integration and validation path.
Configuration order
- Enter the API key and required Base URL.
- Run the connection test.
- Confirm that the account can use the selected model.
- Submit a minimal request using a default size.
- Add reference inputs and advanced parameters only after the basic request succeeds.
Default routing and explicit selection
Specify a provider or model when predictable output matters. Automatic routing can suggest a path but cannot replace model-capability checks.
When one provider fails
- Check the connection test, quota and balance.
- Confirm that the model name is still valid.
- Check that the size and input fields are supported.
- Do not modify unrelated working providers to fix one route.