Install or upgrade to StreamDiffusionTD v0.4.0

Full page: Installation Guide

v0.4.0 is a stack bump (torch 2.11, TensorRT 10.16, CUDA 12.8 only, cuda-link). ControlNet, IP Adapter, StreamV2V, and live LoRA weight now run together on TensorRT.

Fresh local install

1. Python 3.11.9 on PATH. Not 3.12, not 3.13, not 3.10
2. Git on PATH
3. Drag streamdiffusiontd_0.4.0.tox into TouchDesigner
4. Install page: new empty Basefolder (do not name it streamdiffusion)
5. Step 1 Clone, Step 2 Install (venv), Step 3 Install TensorRT
6. Restart TouchDesigner so cuda-link sees CUDALINK_LIB_PATH
7. Acceleration = tensorrt. Start stream. First engine build takes a while
8. python -m sd_installer verify from the installer folder after activating the venv

Daydream / hosted: drag and drop, same as v0.3.1. No local install.

Upgrade from v0.3.1

1. Save the project
2. Load the v0.4.0 tox
3. Install Step 1, take Full Update (repo + fresh venv). Old venv is renamed, not deleted
4. Restart TouchDesigner
5. python -m sd_installer verify

Upgrade from a v0.4.0 beta

Same as v0.3.1, plus delete engines/td/ before first run.

If you used Local ↔ Daydream swap on a beta, reset the tox registry first. Skip this if you never ran a 0.4.0 beta.

1. Close TouchDesigner
2. In File Explorer, paste %APPDATA%\dotsimulate_TD_Tools into the address bar
3. Delete streamdiffusiontd_toxes.json
4. In toxes, delete streamdiffusiontd_Local_0.4.0b.tox and streamdiffusiontd_Daydream_0.4.0b.tox if they are there. Leave 0.3.1 files alone
5. Open TouchDesigner, drop streamdiffusiontd_0.4.0.tox, let it finish loading

First ControlNet stream rebuilds its engine once. Expected.

If Full Update is refused

That checkout's git origin is a personal fork. Use Fresh Install (New Location).

After install

• First ControlNet stream rebuilds once
• TensorRT is required for ControlNet, IP Adapter, and StreamV2V
• Slow installs mentioning pypi.ngc.nvidia.com: see the NameResolutionError FAQ. Updating StreamDiffusionTD does not remove that leftover pip config

Paste python -m sd_installer diagnose in Discord #troubleshoot-central and tag @dotsimulate.