Provider guide ยท Stable

OpenAI-compatible image servers

Bring your image server or proxy to SDAI on Android and iOS.

What this source supports

OpenAI-compatible appears immediately after Swarm UI in the default provider list. It is marked Stable on Android (Google Play, full and FOSS builds) and iOS. The server must implement the OpenAI image generation API; compatibility with chat endpoints alone is not enough.

  • Text-to-image with an editable model ID and positive width and height.
  • Sequential batches, with one image requested at a time.
  • Base64 image data or a downloadable HTTP(S) image URL.
  • Shared gallery, image details and the app's save/export flows.

Img2img, image editing, negative prompts, samplers, sampling steps, CFG scale, seed and quality options are not sent by this source. The server determines available models, supported sizes and account limits. The official OpenAI source has separate configuration.

Connect your server

  1. Start the image server or proxy and enable access from your phone's network.
  2. Open SDAI Configuration and choose OpenAI-compatible.
  3. Enter the complete API prefix, such as http://192.168.1.10:10531/v1 or https://images.example.com/proxy/v1. Include the version and any reverse-proxy path your server requires.
  4. Leave API key empty for a keyless server. Otherwise enter its bearer token.
  5. Tap Connect.
  6. Enter an image model ID available on that server, choose supported dimensions, write a prompt and tap Imagine.

For a server on a computer in the same LAN, enter the computer's LAN IP address. The server must listen on that network interface and its port must be reachable through the firewall. The phone's localhost or 127.0.0.1 refers to the phone. Use HTTPS outside a trusted local network.

Do not paste a generation endpoint into the API base field. SDAI appends models for the connection check and images/generations for generation. It does not add /v1 automatically.

Using an external ChatGPT proxy

openai-oauth is one example of a separately running image proxy. Follow its own setup instructions, complete ChatGPT sign-in there, then enter its reachable API prefix in SDAI. This OpenAI-compatible source does not perform ChatGPT OAuth or run the proxy on your phone. For direct mobile sign-in, use the separate ChatGPT OAuth (Beta) provider.

Use a model your account can access, such as gpt-image-2 when the proxy exposes it. Model access and limits depend on the upstream service. Protect access to a keyless proxy: anyone who can reach its API can use the connected account.

Troubleshooting

Connection is unverified

Some image servers do not expose model discovery. If /models returns 404 or 405, SDAI saves the connection with an unverified notice. Enter the model ID manually. Even a successful connection check does not prove image generation works.

Connection or generation fails

Check the API prefix, network access, authentication, model ID, dimensions and account limits. A server that accepts text chat may still lack image generation. A returned image URL must be reachable from the phone; SDAI does not attach your provider bearer key to image downloads.

The result has a different size

The server may return different dimensions from those requested. SDAI records the decoded image's actual size in gallery details. Unsupported diffusion metadata is not supplied by this source.

For the precise server integration contract, see network/openai-compatible.openapi.json in the source repository. See also the FAQ and Privacy Policy.