invoke-openai-proxy

OpenAI-compatible image API proxy for InvokeAI. Single static Go binary with an embedded admin web UI.

Allows tools like Open-WebUI and any other OpenAI Image API client to generate images via a local InvokeAI installation.

Features

• OpenAI Image API — /v1/images/generations, /v1/images/edits, /v1/images/variations
• Model registry — map OpenAI model names to InvokeAI workflows with parameter injection
• Admin UI — embedded HTMX interface for managing workflows, models, and testing
• Single binary — no dependencies, no database, no CGO
• Real-time — Socket.IO integration for instant completion detection

Quickstart

# Download from releases (or build from source)
./invoke-openai-proxy \
  --listen-ip 0.0.0.0 \
  --port 8080 \
  --invoke-url http://localhost:9090

# Visit http://localhost:8080/admin
# 1. Upload a workflow JSON exported from InvokeAI
# 2. Create a model mapping (set prompt/seed/size field paths)
# 3. Configure your OpenAI client to use http://<host>:8080/v1

Or skip the manual mapping entirely — Quick Setup installs a ready-made workflow preset for SDXL, Flux, Flux2 Klein, Z-Image, SD 1.5, Krea-2, Anima or Qwen Image against a model of your choice:

Open-WebUI Configuration

Set OpenAI API Base URL to http://<proxy-host>:8080/v1 and the models from your registry will appear in the image generation dropdown.

Build from Source

git clone https://github.com/Pfannkuchensack/openaiapi2invokeai-go.git
cd openaiapi2invokeai-go
make build        # → bin/invoke-openai-proxy
make cross        # → linux/amd64, linux/arm64, darwin/arm64, windows/amd64

Requires Go 1.22+.

Configuration

All options available as CLI flags, environment variables, or config.toml:

Flag	Env	Default	Description
--listen-ip	PROXY_LISTEN_IP	127.0.0.1	Bind address
--port	PROXY_PORT	8080	Listen port
--invoke-url	INVOKE_URL	http://127.0.0.1:9090	InvokeAI base URL
--data-dir	PROXY_DATA_DIR	~/.invoke-openai-proxy	Storage for workflows + registry
--api-key	PROXY_API_KEY	(empty)	Bearer token for API auth
--admin-user	PROXY_ADMIN_USER	(empty)	Basic-auth user for /admin
--admin-pass	PROXY_ADMIN_PASS	(empty)	Basic-auth password for /admin
--timeout	PROXY_TIMEOUT	300s	Max wait per generation
--log-level	PROXY_LOG_LEVEL	info	debug/info/warn/error

Priority: flag > environment > config file > default.

Config file location: <data-dir>/config.toml

API Endpoints

GET /v1/models

Lists all registered models in OpenAI format.

POST /v1/images/generations

{
  "model": "sdxl-fast",
  "prompt": "a beautiful sunset over mountains",
  "size": "1024x1024",
  "n": 1,
  "response_format": "b64_json"
}

POST /v1/images/edits

Multipart form: image, mask (optional), prompt, model, size, n

POST /v1/images/variations

Multipart form: image, model, size, n

GET /healthz

Health check endpoint.

Workflow Files

Two JSON shapes are accepted and both are detected automatically on upload:

• Editor export — what Download Workflow in the InvokeAI workflow editor produces (nodes is an array, field values live under data.inputs). It is converted to an API graph the same way the InvokeAI frontend does before enqueueing: notes/current-image nodes and collapsed edges are dropped, and fields fed by an edge stay unset.
• API graph — the raw enqueue_batch graph (nodes is an object keyed by node ID).

Either way the field paths used in the model registry refer to the graph form, i.e. nodes.<node-id>.<field>.

Model Registry

Models are defined in <data-dir>/registry.json. Each model maps an ID to a workflow file and defines which graph nodes receive which parameters:

The admin UI edits this registry directly, so the JSON below is what the table above writes:

{
  "models": [{
    "id": "sdxl-fast",
    "workflow": "sdxl-workflow.json",
    "mapping": {
      "prompt": "nodes.<node-uuid>.value",
      "negative": "nodes.<node-uuid>.value",
      "seed": "nodes.<noise-uuid>.seed",
      "width": "nodes.<noise-uuid>.width",
      "height": "nodes.<noise-uuid>.height",
      "steps": "nodes.<denoise-uuid>.steps",
      "cfg": "nodes.<denoise-uuid>.cfg_scale"
    },
    "defaults": {"steps": 20, "cfg": 7.5},
    "size_presets": {
      "1024x1024": {"width": 1024, "height": 1024},
      "1792x1024": {"width": 1792, "height": 1024}
    }
  }]
}

Use the Admin UI → Workflows → Inspect page to find the correct node UUIDs and field names.