Questions

Docs · Questions

StreamDiffusionTDUsage

What changed across StreamDiffusionTD v0.2.99, v0.3.0, and v0.3.1?

StreamDiffusionTD Version Progression: v0.2.99 to v0.3.1

v0.3.1 (March 24, 2026) - Stability + Creative Tools

Polish update following the v0.3.0 rebuild. Major additions:

  • Built-in FX Processors: Feedback Loop (zoom, pan, rotation) and Feedback Grade (color grading, hue drift). Both run in the image_pre stage, compounding through the diffusion feedback path
  • StreamV2V / Cached Attention: Video-to-video mode is back (was in v0.2.99, missing in v0.3.0). Requires TRT, resolution locked to engine build
  • Installer CLI: verify (13-point check), repair (auto-fix dependencies), diagnose (full system dump for bug reports)
  • Depth TRT Auto-Build: Depth ControlNet TRT engine builds automatically on first use (~60% faster than PyTorch depth)
  • User/Dev Tox Mode: Preserve internal operator modifications across updates
  • FPS Limiting and Benchmark: Control GPU load, one-click benchmark to clipboard
  • Scheduler/Sampler Selection: Choose noise scheduler and sampler
  • Auto-Update System: Check for operator updates from the About page
  • Daydream Cloud Operator: Separate bundled component, pinned to v0.2.2
  • Dozens of bug fixes (prompt weight edge case, FX temporal flickering, multi-ControlNet routing, TRT build crashes, SDXL CN TRT)

v0.3.0 (November 6, 2025) - Complete Rebuild

Complete rebuild from the ground up:

  • SDXL Support: Full SDXL model support
  • IP Adapter: Built-in IP Adapter
  • TensorRT for Local: Previously cloud-only
  • Flexible TRT Engines: Resolution (384-1024) and steps (1-4) changeable without rebuild
  • Daydream Cloud Backend: No-install hosted option

v0.2.99 (August 21, 2025) - Cloud + V2V

  • Built-in cloud backend (Daydream API)
  • V2V temporal consistency (now available again in v0.3.1)
  • 5xxx GPU support
  • Multi-ControlNet + TRT for cloud

Feature Comparison

Feature v0.2.99 v0.3.0 v0.3.1
SDXL models No Yes Yes
IP Adapter No Yes Yes
StreamV2V Yes No Yes (TRT required)
FX Processors No No Yes
Local TensorRT No Yes Yes
Flexible TRT resolution No Yes Yes
Installer CLI No No Yes
Auto-Updates No No Yes
Daydream Cloud Yes Yes Yes

Upgrading

From v0.3.0 to v0.3.1:

  1. Load the new v0.3.1 TOX
  2. Run Step 1 on the Install page to pull code updates
  3. Activate venv and run python -m sd_installer verify
  4. If issues found, run python -m sd_installer repair

From v0.2.99: Install v0.3.1 in a separate folder. Do not overwrite your v0.2.99 installation.


Latest version: v0.3.1 (March 24, 2026). See What’s New for the full changelog.