---
title: StreamDiffusion Visual Engine
emoji: 🎨
colorFrom: purple
colorTo: indigo
sdk: docker
app_port: 7860
hardware: a10g-small
pinned: false
---

# StreamDiffusion Real-Time Visual Engine

WebSocket server for real-time AI visual generation during immersive healing sessions.

## Protocol

**WebSocket endpoint:** `wss://<space-url>/ws`

**Client → Server (JSON):**
```json
{
  "prompt": "sacred geometry, ethereal light",
  "amplitude": 0.3,
  "beat": false,
  "phase": 1
}
```

**Server → Client:** Binary JPEG frame

## Configuration

| Env Var | Default | Description |
|---------|---------|-------------|
| `MODEL_ID` | `stabilityai/sd-turbo` | Diffusion model |
| `TINY_VAE_ID` | `madebyollin/taesd` | Fast VAE decoder |
| `WIDTH` | `512` | Output width |
| `HEIGHT` | `512` | Output height |
| `NUM_STEPS` | `1` | Inference steps (1 = fastest) |
| `GUIDANCE_SCALE` | `0.0` | CFG scale (0 for SD-Turbo) |
| `JPEG_QUALITY` | `75` | Output JPEG quality |