xDiT
xDiTxDiT is an inference engine that spreads image and video diffusion transformers across several GPUs or machines so that large models generate faster. It combines sequence parallelism, PipeFusion pipelining, classifier-free-guidance parallelism and a parallel VAE with caching and compilation speedups, and wraps Diffusers pipelines for FLUX, Qwen-Image, Wan, HunyuanVideo, LTX and other families. It installs from PyPI as xfuser.
Openness
5 high confidence- license
- Apache-2.0 (OSI. LICENSE.txt is the Apache License 2.0 text)
- source
- public (xdit-project/xDiT)
- core features withheld
- no — no ee, enterprise, commercial or proprietary path in the file list. The README names TACO-DiT, a closed ComfyUI variant built on xDiT, as a separate commercial project
xDiT is released under the Apache License 2.0 with its full source public. The project's commercial work, a closed ComfyUI variant called TACO-DiT, is a separate product built on top of the engine rather than a paid edition of it.
- https://raw.githubusercontent.com/xdit-project/xDiT/main/LICENSE.txt recorded 2026-11-12
LICENSE.txt body: "Apache License / Version 2.0, January 2004", the stock text through "limitations under the License."
- https://raw.githubusercontent.com/xdit-project/xDiT/main/README.md recorded 2026-11-12
README: "Using our commercial project **TACO-DiT**, a close-sourced ComfyUI variant built with xDiT"; install with "pip install xfuser" or "pip install -e ." from the repository.
- https://ungh.cc/repos/xdit-project/xDiT/files/main recorded 2026-11-12
Full file list of the main branch, 652 files under xfuser/, examples/, entrypoints/, docs/ and tests/: no ee/, enterprise/, commercial/ or proprietary/ path.
Adoption
2 high confidenceAdoption is measured as PyPI downloads of xfuser, the package the README installs and whose project page links back to the repository. A download count includes automated installs, so it measures use only roughly.
- https://pypi.org/pypi/xfuser/json recorded 2026-11-12
PyPI xfuser 0.7.0, uploaded 2026-09-25: summary "A Scalable Inference Engine for Diffusion Transformers (DiTs) on Multiple Computing Devices", Homepage https://github.com/xdit-project/xDiT.
- https://pypistats.org/api/packages/xfuser/recent recorded 2026-11-12
pypistats for xfuser: last_month 22682.
Capability
3 high confidencexDiT runs both image and video diffusion models across many families, as stable-diffusion.cpp does. It runs no audio or 3D models and composes no workflows, which keeps it below ComfyUI.
- https://raw.githubusercontent.com/xdit-project/xDiT/main/README.md recorded 2026-11-12
README Supported DiTs table: video models including StepVideo, HunyuanVideo, HunyuanVideo-1.5, CogVideoX, Wan2.1, Wan2.2, LTX-2, LTX-2.5 and MiniMax-H3, and image models including Flux, Flux 2, Flux 3 Image klein, Qwen Image, Z-Image Turbo, Stable Diffusion 3, PixArt-Sigma, SANA and HunyuanDiT; parallel methods USP, PipeFusion, CFG parallel and parallel VAE.
Verified 2026-11-12