werk1112

Open WebUI integration

Open WebUI can use Werk1112 as an OpenAI-compatible chat service and as an image generation backend. For images, its OpenAI engine is the preferred integration. Werk also exposes a deliberately small AUTOMATIC1111 compatibility surface for clients that require that protocol.

Start Werk

Generate a key and select a concrete local image model:

werk auth api-key generate
werk serve --host 0.0.0.0 --image-model IMAGE_MODEL

Bind to 0.0.0.0 only when another environment, such as a container, must reach Werk. It exposes the port on every host interface; use host firewall rules and a trusted network. When both applications run directly on the same host, keep the default loopback binding instead:

werk serve --image-model IMAGE_MODEL

Chat connection

Configure an OpenAI-compatible connection in Open WebUI with:

Base URL: http://WERK_HOST:11434/v1
API key:  <generated Werk API key>

The /v1 suffix is required for OpenAI clients. Installed Werk text models are returned by GET /v1/models; whether an individual model can execute depends on its declared task and an available backend.

Werk implements a chat-completions subset rather than the complete OpenAI API. In particular, automatic tool-call orchestration and the Responses API are not provided by werk serve.

Recommended image engine: OpenAI

Configure Open WebUI’s image generation settings or equivalent environment variables as follows:

ENABLE_IMAGE_GENERATION=true
IMAGE_GENERATION_ENGINE=openai
IMAGES_OPENAI_API_BASE_URL=http://WERK_HOST:11434/v1
IMAGES_OPENAI_API_KEY=<generated Werk API key>
IMAGE_GENERATION_MODEL=gpt-image-1
IMAGE_SIZE=1024x1024

Typical host values are:

• 127.0.0.1 when both processes run directly on the same host;
• host.docker.internal when Open WebUI runs in Docker and Werk runs on the host, where that hostname is provided by the container runtime;
• a deliberately reachable host or VM address when the applications run in separate environments.

gpt-image-1 is an alias resolved to the model selected with werk serve --image-model IMAGE_MODEL. An explicit installed Werk model ID can be used instead. Werk returns embedded Base64 by default, so Open WebUI does not need to make a second authenticated output request.

The current boundary is text-to-image generation. Open WebUI’s OpenAI multipart image-edit request is not supported by Werk’s JSON image-edit endpoint. Image streaming, partial image events and transparent-background enforcement are also not implemented.

Alternative image engine: AUTOMATIC1111

Use this only when the client must speak the A1111 protocol:

ENABLE_IMAGE_GENERATION=true
IMAGE_GENERATION_ENGINE=automatic1111
AUTOMATIC1111_BASE_URL=http://WERK_HOST:11434
AUTOMATIC1111_API_AUTH=werk:<generated Werk API key>

The A1111 base URL is the server root and therefore does not include /v1. Werk accepts Basic authentication with username werk for these compatibility routes. It implements txt2img, model listing, options and coarse progress only; it is not an A1111 extension/script runtime. See the AUTOMATIC1111 integration guide for the precise subset.

Browser and container notes

Open WebUI normally calls Werk from its server process, so browser CORS does not participate. If a browser frontend calls Werk directly, start Werk with the exact trusted origin, for example:

werk serve --image-model IMAGE_MODEL \
  --cors-origin http://127.0.0.1:3000

Wildcard, null and file:// origins are rejected. CORS does not replace API key authentication.

For the underlying request fields, output lifetime and compatibility limits, see the HTTP API reference.

This site is open source. Improve this page.