WebUI-Claw

OpenClaw + Stable Diffusion WebUI integration for mobile image generation.

Users can send natural-language prompts from phone chat apps, and OpenClaw routes requests to WebUI API and returns generated images back in chat.

Example: Generate 10 images: cyberpunk cat detective, cinematic lighting, 768x1024, steps=30, cfg=7

---

Telegram demo screenshots

Real chat screenshots from Telegram flow (generate batch images and return best picks):

---

Why this project

• Mobile-first image generation workflow
• Natural-language parameter parsing (count, size, steps, cfg)
• Works across channels supported by OpenClaw
• Easy self-hosted deployment with Docker Compose

---

Multi-channel support (not Telegram-only)

This project is channel-agnostic because OpenClaw handles messaging.

If OpenClaw supports a channel, this flow can work there too (with channel-specific media limits):

• Telegram
• WhatsApp
• Discord
• Slack
• Signal
• Line
• iMessage
• IRC / Google Chat (depending on OpenClaw setup)

So yes — once integrated with OpenClaw, users can use other mobile apps, not only Telegram.

---

Architecture

flowchart LR
  A[Mobile Chat App] --> B[OpenClaw]
  B --> C[WebUI-Claw skill]
  C --> D[AUTOMATIC1111 /sdapi/v1/txt2img]
  D --> C --> B --> A

Loading

---

---

Repository structure

.
├─ README.md
├─ README.zh-CN.md
├─ .env.example
├─ docker-compose.yml
├─ docs/
│  └─ GITHUB_DEPLOY_CN.md
├─ scripts/
│  ├─ deploy.sh
│  └─ healthcheck.sh
└─ skill/
   ├─ SKILL.md
   └─ scripts/
      ├─ generate.py
      └─ parse_and_generate.py

---

Quick Start

git clone https://github.com/YoujunZhao/WebUI-Claw.git
cd WebUI-Claw
cp .env.example .env
bash scripts/deploy.sh

Default API endpoint after startup:

• http://127.0.0.1:7860/sdapi/v1/txt2img

---

Connect skill to OpenClaw

mkdir -p ~/.openclaw/workspace/skills/openclaw-webui-image
cp -r skill/* ~/.openclaw/workspace/skills/openclaw-webui-image/

Environment variables for OpenClaw runtime:

export SD_WEBUI_URL=http://127.0.0.1:7860
export SD_WEBUI_TIMEOUT=180

---

Mobile command examples

• Generate 1 image: neon cyberpunk city, rainy street, cinematic
• Generate 10 images: Chinese ink landscape, morning mist, 512x768
• Generate 4 images: robotic cat, product white background, steps=30,cfg=7

Chinese works as well:

• 生成10张图：赛博朋克猫咪侦探，电影光影，768x1024，steps=30,cfg=7

---

Built-in parameter parsing

parse_and_generate.py can parse:

• Generate 10 images / 生成10张图 → n_iter=10
• 512x768 / 1024*1024 → width/height
• steps=30 / 步数30 → steps=30
• cfg=7 / cfg 7 → cfg_scale=7

Defaults come from .env if not specified.

---

FAQ

Images are not returned to phone

• Check OpenClaw channel configuration
• Check skill output contains images[] (base64)
• Check media/file-size limits on your channel

WebUI API unreachable

docker compose ps
bash scripts/healthcheck.sh

Slow generation

• Lower resolution or steps
• Use GPU
• Add queue/progress strategy in later versions

---

Security notes

• Do not expose port 7860 to public Internet directly
• Add reverse proxy + auth for remote access
• Avoid logging sensitive prompts permanently

---

License

MIT