vidax
Copyright 2026 vidax authors

This product is licensed under the Apache License, Version 2.0 (see LICENSE),
EXCEPT for the third-party-derived portions identified in section 2 below.

================================================================================
1. Third-party model implementations
================================================================================

vidax re-implements, in JAX/Flax, the architectures of several third-party
PyTorch video diffusion models. This repository contains NO third-party source
code verbatim and NO model weights. Each implementation was written with
reference to the corresponding upstream project's technical report and its
released source code.

Upstream projects and the license of their SOURCE CODE (which is distinct from
the license of their released model weights):

  Wan2.1, Wan2.2 -- Alibaba (Wan team)
      https://github.com/Wan-Video/Wan2.1
      https://github.com/Wan-Video/Wan2.2
      Source code: Apache License 2.0

  CogVideoX / CogVideoX 1.5 -- THUDM / ZhipuAI
      https://github.com/THUDM/CogVideo
      Source code: Apache License 2.0

  LTX-Video (0.9.8) -- Lightricks
      https://github.com/Lightricks/LTX-Video
      Source code: Apache License 2.0

  Cosmos-Predict2.5 -- NVIDIA
      https://github.com/nvidia-cosmos/cosmos-predict2.5
      Source code: Apache License 2.0

  Cosmos 3 (Nano / Edge) -- NVIDIA
      https://github.com/NVIDIA/cosmos
      Source code: OpenMDW-1.1 (https://openmdw.ai/license/1-1/)

  HunyuanVideo, HunyuanVideo-1.5 -- Tencent
      https://github.com/Tencent-Hunyuan/HunyuanVideo
      https://github.com/Tencent-Hunyuan/HunyuanVideo-1.5
      Source code: Tencent Hunyuan Community License Agreement

  LTX-2.5 -- Lightricks
      https://github.com/Lightricks/LTX-2
      Source code: LTX-2.x Community License Agreement

The vidax implementations of Wan, CogVideoX, LTX-Video and Cosmos are provided
under the Apache License 2.0 (this project's license). Please retain this NOTICE
when redistributing.

================================================================================
2. Portions under additional (non-Apache) terms
================================================================================

The Tencent Hunyuan Community License Agreement and the LTX-2.x Community
License Agreement are NOT permissive open-source licenses. Among other terms
they include territorial restrictions (the Tencent Hunyuan Community License
does not apply in the European Union, the United Kingdom, or South Korea) and
acceptable-use and commercial-scale conditions.

To the extent the following parts of vidax are derivative works of the
corresponding upstream source code, your use, modification and redistribution
of THOSE PARTS is additionally subject to the applicable upstream license:

  Tencent Hunyuan Community License Agreement
    (https://github.com/Tencent-Hunyuan/HunyuanVideo/blob/main/LICENSE.txt):
      src/vidax/models/hunyuan_video/**
      src/vidax/translator/mappings/hunyuan_video.py
      src/vidax/translator/mappings/hunyuan_video1_5.py

  LTX-2.x Community License Agreement
    (https://github.com/Lightricks/LTX-2/blob/main/LICENSE):
      src/vidax/models/ltx2_5/**
      src/vidax/translator/mappings/ltx2_5.py

Each of those directories also contains a LICENSE file restating this. Refer to
the upstream repositories for the full, current license text. If you cannot
accept those terms, do not use the parts listed above; the rest of vidax
remains available to you under the Apache License 2.0.

================================================================================
3. Model weights
================================================================================

vidax distributes no model weights. Every checkpoint you obtain from a third
party is governed by that party's own license -- see the References table in
README.md. You are responsible for complying with the license of any weights
you download or use.
