Our take
Inactiveno commits in over a year- Activity
- 19
- Community
- 70
- Issues
- 21
- Pull requests
- 20
+15 stars in the last 30 days
StreamDiffusion is a specialized starting point for GPU-backed interactive diffusion, but its lack of recent maintenance makes production adoption a decision that should include a compatibility and support plan.
Good fit if
- You already use Python, PyTorch, and Diffusers and can run the workload on a suitable GPU.
- You want to experiment with streaming image generation or live image-to-image input, and can verify the results on your own setup.
- Apache-2.0 licensing fits your project, subject to reviewing the license and any separate model terms.
Look elsewhere if
- You need an actively maintained dependency: there have been no commits in over a year and no release in the past year.
- Your project cannot tolerate stalled issue handling: no issues were closed in the last 90 days, and open issues remain.
- You need generation without a GPU-oriented PyTorch setup or do not want to handle TensorRT installation and engine building for that acceleration path.
All health signals
| Last commit | 2024-12-04 (1.8 years ago) |
|---|---|
| Commits, last 90 days | 0 |
| Releases, last 12 months | 0 (latest v0.1.1, 2023-12-30) |
| Contributors | 23 (top contributor: 26% of commits) |
| Issues closed, last 90 days | 0 |
| Project age | 2 years |
Checked on 2026-10-07 with the GitHub API.
Project history
Daily GitHub snapshots since Oct 7, 2026. History covers stars and forks; health signals are not tracked historically.
Stars
10,835Forks
832Change since Nov 14, 2026: +1 stars·-1 forks
Overview
StreamDiffusion is a Python library that modifies diffusion pipelines to generate images interactively, rather than treating each image as a separate, slow request. It is aimed at developers who want to build responsive text-to-image or image-to-image experiences, including workflows driven by live webcam or screen-capture input.
Its approach combines batched processing and caching with techniques for reducing redundant work. The repository includes examples and browser demos, while the reported speed figures are tied to specific models and an RTX 4090 system, not a general performance guarantee.
Key Features
- Wraps Diffusers Stable Diffusion pipelines for streaming text-to-image and image-to-image generation.
- Uses Stream Batch processing to handle generation in batches.
- Offers Residual Classifier-Free Guidance modes intended to reduce the cost of guidance.
- Provides a Stochastic Similarity Filter that can skip processing when successive input frames are similar.
- Includes input and output queues and pre-computation for KV caches.
- Supports acceleration with TensorRT, which requires an additional extension and engine-building step.
- Provides example workflows and real-time text-to-image and webcam or screen-capture image-to-image demos.
Use Cases
- A creative-tool developer prototyping a live text-to-image interface where prompts should produce a continuing stream of results.
- A computer-vision or generative-media developer transforming webcam or screen-capture frames into stylized images in real time.
- A team experimenting with interactive installations or other GPU-powered visual experiences, when it can validate performance on its own hardware and models.
- A Diffusers user exploring pipeline-level inference optimizations before building a custom streaming generation workflow.
What you need
Detected in the repository
- Python (from pyproject.toml)
- A Dockerfile, so it can run in a container
- Automated checks on GitHub Actions
License in plain words
Apache-2.0permissive
- Commercial use: yes
- Modify and redistribute: yes
- You must keep: the license, the NOTICE file and a note of your changes
- Share your changes: no
- Includes an explicit patent grant from the contributors:
A summary, not legal advice: the LICENSE file is what applies.
Getting Started
The README recommends Python 3.10 and provides PyTorch installation instructions for CUDA systems. After setting up the appropriate PyTorch build, install the latest repository version with:
pip install 'git+https://github.com/cumulo-autumn/StreamDiffusion.git@main#egg=streamdiffusion[tensorrt]'
See the README for environment setup, TensorRT installation, examples, and platform-specific notes.
Alternatives
- Open-Higgsfield-AI: Open-Higgsfield-AI is a JavaScript studio for self-hosted image and cinematic workflows; image generation is currently paused, unlike StreamDiffusion's generation focus.
| Project | Language | License | Stars | Status |
|---|---|---|---|---|
| StreamDiffusion | Python | Apache-2.0 | 10.8k | Inactive |
| Open-Higgsfield-AI | JavaScript | – | 180 | Quiet |
Considerations
- The repository is marked inactive by its recent activity: no commits in over a year, no recent releases, and no issues closed in the last 90 days. It is not archived, and contributions come from multiple people rather than being concentrated in one contributor, but there is no recent maintenance signal.
- The documented setup centers on Python, PyTorch, CUDA, and GPU inference. TensorRT acceleration adds installation and engine-building work; confirm compatibility with your hardware and dependencies before relying on it.
- The README's FPS results are measurements from a specified RTX 4090 system with particular models and settings. They should not be treated as expected performance on other configurations.
- The repository is licensed under Apache-2.0. Check the license terms and the separate licenses or usage terms for any diffusion models and LoRAs you choose to use.
Found this useful?
Share it with someone who would like StreamDiffusion.