Docs · Questions
StreamDiffusionTDInstallation
What are the common installation and setup issues with StreamDiffusionTD, and how do I fix them?
StreamDiffusionTD Installation Issues - Solutions
First Step: Use the Installer CLI (v0.3.1+)
v0.3.1 includes a built-in installer CLI. Activate the venv from the Install page and run:
cd StreamDiffusion-installer
python -m sd_installer verify # 13-point environment check
python -m sd_installer repair # auto-fix common dependency issues
python -m sd_installer diagnose # full system info for bug reports
The repair command fixes most common issues automatically (onnx version, diffusers fork, protobuf pin).
Pre-Installation Checklist
- Python Version: Must be 3.11.9 or 3.10.9 (NOT 3.12, NOT 3.13)
- CUDA Version: 12.8 recommended
- Install Location: Use a NEW folder (not over previous versions)
- Folder Name: Don’t name it “streamdiffusion” (causes module conflicts)
- TouchDesigner: 2025 version recommended
Installation Strategy
- Install in a NEW folder separate from any existing installation
- Use the “Base Folder” setting in the Install tab to choose a new location (e.g.,
C:/AI/StreamDiffusion_v031/) - The operator downloads the exact pinned version from the version manifest
- On startup, the operator verifies your local repos match the expected version
Common Installation Errors
1. TensorRT Installation Failures
- Error: Dependency conflicts during pip installation
- v0.3.1 installs TRT sub-packages (
tensorrt_cu12_libs+tensorrt_cu12_bindings) instead of the broken meta-package - If you still hit issues, run
python -m sd_installer repair
2. “Acceleration Has Failed” Error
- Symptoms: Stream won’t start, loops between “server connected” and “server stopped”
- Fixes: Delete engines at
StreamDiffusion/engines/td/, ensure IP Adapter is OFF for sd-turbo, restart stream
3. “polygraphy” Module Missing
- Fix: Click the “Install TensorRT” button on the Install page
4. Module Import Errors
- Error:
Import failed: no module named 'streamdiffusion.config' - Fix: Install in a separate folder with a name other than “streamdiffusion”
5. “cudart” Import Error (50-series GPUs)
- Fix: Ensure CUDA 12.8 is installed, restart computer
6. flash-attn Crash
- Fix:
pip uninstall flash-attnin your SD venv
7. v0.3.0 Fix Scripts Breaking v0.3.1
- Running the v0.3.0
Fix_All_Dependencies.batoverwrites the required diffusers fork - Fix:
python -m sd_installer repair
Working Solutions
Primary Fix:
- Verify Python 3.11.9 with
python --version - Fresh install in NEW folder
- Select CUDA 12.8 during installation
- Initial test config: sd-turbo, 512x512, IP Adapter OFF, ControlNet weight 0, TensorRT ON
- Run
python -m sd_installer verifyto confirm everything is correct - Let TensorRT build (20-30 minutes first time, don’t interrupt)
Dependency Pins (v0.3.1)
| Package | Requirement | Why |
|---|---|---|
| numpy | < 2.0 | SD and mediapipe break on 2.x |
| protobuf | 4.25.8 | 6.x breaks mediapipe |
| onnx | 1.18.0 | >= 1.20 crashes engine builds |
| peft | >= 0.6.0 | Required for StreamV2V |
| diffusers | varshith15 fork | Must be the fork, not vanilla |
GPU Compatibility Status
| GPU Series | Status | Notes |
|---|---|---|
| 50xx (5070, 5080, 5090) | Works | CUDA 12.8 required |
| 40xx (4090, 4080) | Works | Most reliable |
| 30xx (3070, 3050) | Works | Lower FPS, use sd-turbo |
| Mac | Limited | Cloud mode only recommended |
Still having issues? Post in Discord #troubleshoot-central with output from python -m sd_installer diagnose