English | 中文

@imcaptor/openclaw-poe-image

OpenClaw plugin for image generation via Poe API.

Supports multiple models through a single API key: nano-banana-2, flux-2-pro, seedream-4.5, gpt-image-2, gpt-image-1.5, and more.

Install

openclaw plugins install @imcaptor/openclaw-poe-image

Setup

Set your Poe API key:

export POE_API_KEY="your-key-here"

Or configure via OpenClaw:

openclaw config set plugins.entries.poe-image.config.poeApiKey "your-key-here"

Privacy Notice

When generating images, this plugin sends your prompt and any reference images you provide to the Poe API. Reference images are included as data: URIs in the request. Do not upload sensitive or private images, and prefer a revocable, least-privilege Poe API key.

Usage

Once installed, the image_generate tool automatically includes Poe as an available provider. Ask your agent to generate images:

Generate an image of a friendly lobster mascot using poe/nano-banana-2

Or set Poe as the default image provider:

{
  agents: {
    defaults: {
      imageGenerationModel: {
        primary: "poe/nano-banana-2",
      },
    },
  },
}

Or use the CLI:

openclaw config set agents.defaults.imageGenerationModel.primary "poe/nano-banana-2"

Available models

Model	Speed	Quality	Cost	Best for
nano-banana-2	Fastest	Good	Cheapest	Quick tests, batch generation
nano-banana	Very fast	Good	Cheap	Fast iteration
nano-banana-pro	Medium	Excellent	Moderate	High quality generation
flux-2-pro	Slow	Ultra-high	Expensive	Photorealistic images
flux-kontext-pro	Slow	Ultra-high	Expensive	Layout control, reference-guided
flux-schnell	Fast	Very good	Moderate	Balanced quality and speed
seedream-4.0	Medium	Good	Cheap	Chinese cultural art, Chinese prompts
seedream-4.5	Medium	Very good	Moderate	Chinese text rendering
seedream-5.0-lite	Fast	Good	Cheap	Fast generation
gpt-image-2	Medium	Excellent	Expensive	Latest OpenAI model, instruction following and text rendering
gpt-image-1.5	Medium	Excellent	Expensive	Instruction following, text rendering
gpt-image-1	Medium	Very good	Moderate	General image generation

Specify a model

Generate a photorealistic portrait using poe/flux-2-pro

Image editing

Some models (gpt-image-2, gpt-image-1.5, gpt-image-1, flux-kontext-pro) support reference image editing:

Make this photo look like a watercolor painting

Multi-image input (image-to-image / multi-image fusion)

Up to 3 reference images can be supplied in a single request. Models like gpt-image-2, gpt-image-1.5 and gpt-image-1 are particularly good at fusing multiple references into one output. Refer to them by order in the prompt:

Put the girl from image 1 in the dress from image 2 and the pose from image 3

Different models accept different numbers of references (e.g. flux-kontext-pro only accepts 1). The plugin allows up to 3 across the board and lets Poe return the model-specific error when the limit is exceeded.

Local development & build

npm install
npm run build   # emits to dist/
npm test

Starting from 1.2.0, OpenClaw 2026.5.4+ requires npm/ClawHub plugin packages to ship a compiled runtime output (dist/index.js). The package.json declares openclaw.runtimeExtensions, and npm publish runs prepublishOnly to build it for you.

Changelog

• 1.2.2 — Fixes README language switch links on ClawHub by using stable GitHub URLs instead of relative paths.
• 1.2.1 — Uses English as the default README for GitHub and ClawHub, with a separate Chinese README available from the language switcher.
• 1.2.0 — Compatible with OpenClaw 2026.5.4+. Ships compiled dist/ runtime output, fixing the requires compiled runtime output for TypeScript entry ./index.ts install error.
• 1.1.1 — Legacy release, works on OpenClaw <= 2026.5.3 only.

License

MIT