Installation Guide

Choose Your Mode

Daydream Cloud (Recommended for Quick Start)

Zero installation required. Drag and drop the TOX and start streaming immediately.

Local Installation (Full Features)

Run on your own GPU with full control over all features including FX Processors, StreamV2V, and TensorRT acceleration.

Daydream Cloud Installation

No Python, CUDA, or GPU required.

Steps

  1. Load Operator: Drag StreamDiffusionTD.tox into TouchDesigner
  2. Install Page: Select “Daydream” in backend selection
  3. API Key: Enter your Daydream API key
  4. Start Stream: Ready to use immediately

Daydream Features

  • SDXL-Turbo and SD-Turbo models
  • Multiple ControlNets (OpenPose, HED, Canny, Depth, Color)
  • IP Adapter with FaceID
  • Works on Mac and Windows
  • No external window (WebRender TOP)

Note: FX Processors, StreamV2V, and custom processor support are only available with the Local backend.

Local Installation (v0.3.1)

Prerequisites

Python 3.11.9 (Required)

  • Recommended version: 3.11.9 (3.10.9 also works)
  • Check “Add to PATH” during installation
  • NOT compatible: Python 3.12, 3.13

Git: Download here - check “Add to PATH”

Important: Separate Installation Folder

v0.3.1 MUST be installed in a NEW folder separate from any existing v0.2.99 or v0.3.0 installation.

  • Do NOT upgrade in place from previous versions
  • Do NOT name the folder “streamdiffusion” (causes module conflicts)

Good examples:

  • C:/AI/StreamDiffusion_v031/
  • D:/SD/StreamDiff031/

Installation Steps

  1. Load Operator: Drag StreamDiffusionTD.tox into TouchDesigner
  2. Install Page: Go to the Install parameter page
  3. Set Base Folder: Choose a NEW folder location
  4. Download StreamDiffusion: Click “Clone StreamDiffusion” button. The operator downloads the exact pinned version from the version manifest
  5. Select CUDA Version: Choose CUDA 12.8 when prompted
  6. Install Virtual Environment: Click “Install StreamDiffusion” button, wait for completion
  7. Install TensorRT: Click “Install TensorRT” button (required for ControlNet, IPAdapter, and StreamV2V)

Verify Your Installation

After installation, activate the venv from the Install page and run:

cd StreamDiffusion-installer
python -m sd_installer verify

This runs 13 checks covering PyTorch CUDA, StreamDiffusion core, numpy, diffusers fork, protobuf, onnx, peft, and more. If anything fails, run:

python -m sd_installer repair

The repair command auto-fixes common issues like onnx version conflicts, diffusers fork problems, and protobuf pins.

For bug reports, run:

python -m sd_installer diagnose

This outputs your GPU info (name, VRAM, compute capability), full package version dump, and environment details.

First Run

  1. Go to Models page
  2. Set Acceleration to “tensorrt”
  3. Click Start Stream
  4. Wait for TensorRT engine build (20-30 minutes first time)
  5. Default model: SDXL-Turbo

Initial Test Configuration

For first-time testing, use these settings to avoid issues:

  • Model: stabilityai/sd-turbo (less VRAM than SDXL)
  • Resolution: 512x512
  • IP Adapter: OFF
  • ControlNet: Weight 0 (enabled but not active)
  • Acceleration: TensorRT

Version Tracking

v0.3.1 includes commit tracking. On startup, the operator checks that your local StreamDiffusion and installer repos match the expected versions from the version manifest. If there’s a mismatch, Step 1 on the Install page will show the specific issue, and Steps 2/3 will be disabled until the repos are up to date.

Upgrading from Previous Versions

From v0.3.0

If you were already on v0.3.0:

  1. Load the new v0.3.1 TOX
  2. Go to the Install page. Step 1 should show an X indicating the repos need updating
  3. Click Step 1 to fetch the latest pinned branch and bring your code up to date
  4. Activate venv and run python -m sd_installer verify to confirm everything checks out

From v0.2.99 or Earlier

Fresh installation required:

  1. Keep your existing v0.2.99 installation
  2. Install v0.3.1 in a completely separate folder
  3. v0.3.1 now has StreamV2V (cached attention) back, so you may not need v0.2.99 unless you prefer the non-TRT V2V path

Auto-Updates

v0.3.1 includes an auto-update system. Sign in on the operator’s About page to enable update checks. The operator checks once every 24 hours for new TOX versions and can download and apply updates directly from TouchDesigner.

Common Installation Issues

“polygraphy” Module Missing

Error: ModuleNotFoundError: No module named 'polygraphy' Fix: You skipped step 7. Go back and click “Install TensorRT”.

“cudart” Import Error

Error: cannot import name 'cudart' from 'cuda' Context: Reported on RTX 5070 with CUDA 12.8. Not clearly resolved. Check Discord for updates.

Python Version Mismatch

Error: Installation fails with torch version errors Fix: Check python --version. Must be 3.10.x or 3.11.x. Uninstall Python 3.12/3.13.

Folder Naming Conflict

Error: Import failed: no module named 'streamdiffusion.config' Fix: Don’t name your installation folder “streamdiffusion”. Use a different name.

Acceleration Has Failed

Error: Exception: Acceleration has failed Fix: Delete the engines folder at StreamDiffusion/engines/td/ and restart. Also ensure IP Adapter is OFF for sd-turbo.

flash-attn Crash

Error: Pipeline crashes on import Fix: If you have flash-attn installed, uninstall it: pip uninstall flash-attn in your SD venv.

v0.3.0 Fix Scripts Breaking v0.3.1

Problem: Running the v0.3.0 Fix_All_Dependencies.bat on a v0.3.1 install overwrites the required diffusers fork Fix: Run python -m sd_installer repair to reinstall the correct diffusers fork.

Troubleshooting

See the Troubleshooting Guide for more detailed solutions.