SolarWM/NOTICE at xpu · isl-org/SolarWM · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
SolarWM
Copyright 2026 SolarWM Contributors

Licensed under the Apache License, Version 2.0 (see LICENSE).

--------------------------------------------------------------------------------
Third-party source included in this repository
--------------------------------------------------------------------------------

  SolarRenderer (https://github.com/Enderfga/SolarRenderer), Apache-2.0
    Copyright 2026 SolarRenderer authors.
    Adapted from revision 5a5da36c2cc0b91c00bac85b906cbf7cad469e84 under
    solar/renderer/, with shared observation-packing, attention, LoRA,
    and selective-checkpointing support in the MiniMax-H3 backend.
    The renderer examples, training configurations, and solarrenderer-caption
    skill and helper scripts come from the same release.
    ComfyUI, Reactor, evaluation tools, and expanded documentation are adapted
    from revision e018b0f673de865711fe99844d3976865722b8d0 under
    deploy/solar_renderer/comfyui/, deploy/solar_renderer/reactor/, eval/solar_renderer/,
    and docs/solar_renderer/. Runtime imports, asset paths, deployment inputs,
    and documentation commands are modified for the integrated SolarWM layout.
    Long-video rendering, combined LoRA merging, causal condition positioning,
    long-input encoding, and encoded-silence extension are updated through
    revision 41e136e64b30d4aa0095070ce055ff00ae90fe3b (main, 2026-10-23).
    These updates are adapted under solar/renderer/ in render.py,
    causal.py, preencode.py, and silence.py, with docs/solar_renderer/ updates.
    Renderer-specific causal and silence changes are isolated from the shared
    MiniMax-H3 backend.
    Reactor optimization helpers also accept the merged native model as well
    as the earlier PEFT-wrapped model in this SolarWM adaptation.
    MiniMax-H3 and SolarRenderer model weights are obtained separately and
    remain subject to the MiniMax H3 Community License and Acceptable Use Policy.

  Wan 2.2 (Alibaba Wan Team), Apache-2.0
    Copyright 2024-2025 The Alibaba Wan Team Authors. All rights reserved.
    Vendored under solar/backends/wan22/runtime/modeling/:
      model.py, vae.py, t5.py, attention.py, tokenizers.py
    causal_model.py is a MODIFIED derivative of the above, extending the Wan
    architecture with causal attention and camera conditioning.
    The Chinese negative prompts in the Wan example configs are Wan defaults.

  PRoPE: Projective Positional Encoding for Multiview Transformers, MIT
    Copyright (c) the PRoPE authors.
    Vendored as solar/backends/wan22/runtime/modeling/camera_prope.py
    (dual-licensed Apache-2.0 AND MIT) and adapted in the LTX and H3 backends
    as torch_prope.py.

Each vendored file keeps its upstream copyright header. Files modified by this
project state so in their header.

--------------------------------------------------------------------------------
Backbones and weights
--------------------------------------------------------------------------------

Model weights and backbone packages are NOT distributed here and carry their own
terms, several of which are not Apache-2.0. Obtain each from its publisher and
check its licence before redistributing anything you produce:

  Wan 2.2 TI2V-5B / I2V-A14B          LTX-2.5 (incl. the ltx-core packages)
  MiniMax-H3                          the text encoders each backbone requires

--------------------------------------------------------------------------------
Data
--------------------------------------------------------------------------------

The corpus is produced by the SolarWM Data Engine and annotates public source
datasets that are not redistributed here; each keeps its own terms, several of
which are non-commercial. See the licence and citation registry distributed
with SolarWM-Data.