StreamDiffusionTD
Real-time AI diffusion streaming for TouchDesigner. Current release is v0.4.1, a maintenance update on v0.4.0, the release where ControlNet, IP Adapter, StreamV2V, and live LoRA weight run together on TensorRT.
Key Features
- The combo: ControlNet + IP Adapter + StreamV2V + live LoRA weight, together, on TensorRT
- SDXL Support: Native SDXL-Turbo model support
- IP Adapter / FaceID: Image prompt conditioning, corrected in v0.4.0
- TensorRT 10.16: Local acceleration, CUDA 12.8, 50-series on a production TensorRT
- GPU to GPU transport: cuda-link keeps frames on the GPU between TD and the stream
- Daydream Cloud: Zero installation required for hosted backend (unchanged from v0.3.1)
- ControlNet: Multiple ControlNets, live-feed cache via
Cncachedecay - FX Processors: Feedback loop (persistent canvas) and color grading, plus custom processors
- StreamV2V: Cached attention plus Feature Injection
- Installer: Full Update (repo + fresh venv), plus verify / diagnose / repair
Quick Start
- Installation Guide - Setup for local or cloud
- Parameters Reference - All settings explained
- FX Processors - Built-in effects and custom processing pipeline
- Usage Examples - Essential workflows
- Troubleshooting - Common issues and fixes
System Requirements
Local Mode
- OS: Windows 10/11
- GPU: NVIDIA with 6GB+ VRAM (24GB recommended for all SDXL features)
- Python: 3.11.9. Not 3.12, not 3.13, not 3.10
- CUDA: 12.8 (only option since v0.4.0)
Daydream Cloud Mode
- No installation required - drag and drop the TOX
- Works on any system including Mac
- Internet connection. Login is the Daydream Login pulse on the Install page, not an API key
Version Comparison
| Feature | v0.3.1 | v0.4.x |
|---|---|---|
| ControlNet + IPA + V2V + LoRA together | No | Yes, on TensorRT |
| Live LoRA weight (no rebuild) | No | Yes |
| GPU to GPU transport (cuda-link) | CPU shared memory | Yes |
| ControlNet live-feed cache | Interval was inert on live video | Cncachedecay |
| StreamV2V Feature Injection | Partial | Yes, on by default |
| TensorRT / torch / CUDA | 10.12 / 2.7 / 12.8 | 10.16 / 2.11 / 12.8 only |
| Installer | verify / repair / diagnose | Full Update (repo + fresh venv) |
| Daydream | v0.2.2, drag and drop | Unchanged |
Upgrading from v0.4.0: get the v0.4.1 tox, Install Step 1 shows Update Available, take Full Update. Engines are preserved.
Upgrading from v0.3.1: load the v0.4.1 tox, Install Step 1, take Full Update. Details on the Installation Guide.
Coming from v0.2.99 or v0.3.0: fresh install in a separate folder. Do not upgrade those in place.
Download
Available on Patreon for supporters.
Support
Discord support available via Patreon connected apps or the About Me page in the operator.
Looking for v0.2.99? The legacy version is documented on the Legacy Docs page. StreamV2V is in v0.4.0 (TensorRT required). See What’s New for v0.4.1.