ClearRealityV1 - model license and provenance
================================================

Model            : ClearRealityV1 - SPAN (Swift Parameter-free Attention
                   Network) 4x real-world super-resolution
                   (clearreality-x4.onnx, ONNX, opset 17, FP32)
Upstream source  : https://huggingface.co/Kim2091/ClearRealityV1
                   (ONNX/fp32/4x-ClearRealityV1-fp32-opset17.onnx)
Author           : Kim2091 (the model's author and repository owner)
Architecture     : SPAN - https://github.com/Aising-AI/SPAN
License          : Apache License 2.0 - DECLARED (see "How the licence was
                   established" below; the full Apache-2.0 text is included
                   in this package as LICENSES/Apache-2.0.txt)
Commercial use   : PERMITTED under Apache-2.0 paragraph 2
Redistribution   : PERMITTED with notice retention (Apache-2.0 paragraph 4)
Downloaded       : 2026-10-10
Verified         : the exact bytes shipped here hash to
                   bbce9d5a653281cfae07788d620bd4ec45712709bdf6349a06a893159efd97ce
                   (1,718,947 bytes), which matches the LFS object id
                   published by the upstream repository.
                   onnx.checker validates the graph: 82 nodes, ai.onnx opset
                   17 only, 3-channel RGB in and out, dynamic H and W. Runs on
                   the ONNX Runtime CPU execution provider - no GPU, CUDA or
                   DirectML operator is used anywhere in the graph.

How the licence was established
-------------------------------
The upstream repository carries NO LICENSE file. The Apache-2.0 grant is
recorded in two author-controlled places:

  1. the repository's machine-readable metadata on Hugging Face, which
     declares  license:apache-2.0 ; and
  2. the model card README, which states "License: Apache 2.0".

This package ships the full Apache-2.0 text and attributes the author and
source above. Stated plainly: this is a DECLARATION made by the uploader,
not a licence file published by an independent upstream project, and it is
recorded here as such rather than as an independently audited grant.
Anyone redistributing these weights beyond ImgSera should re-confirm the
terms with the author.

Training data note: the model card states training used the author's own
datasets plus Nomos8k and FaceUp. No claim is made about the terms of those
third-party datasets; that is the author's responsibility as distributor of
the weights, and it is noted here so the question is not lost.

Apache License 2.0 summary (NOT a substitute for the full text):
You may use, reproduce, and distribute the work, commercially or not,
provided you include a copy of the Apache License 2.0 and NOTICE file
contents with any redistribution and state significant changes. This file
serves as that NOTICE for the ImgSera redistribution.
