# AI Upscale Notice

Aivium bundles a local image upscaler (`aivium-upscale` sidecar) built on
Real-ESRGAN and the ncnn inference engine. No data leaves the machine; the
upscaler runs on the GPU through Vulkan or on the CPU.

## Components

| Component                                            | License                                                                                                         | Copyright                                                 | Source                                                                   | Bundled                                                            |
| ---------------------------------------------------- | --------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------- | ------------------------------------------------------------------------ | ------------------------------------------------------------------ |
| `aivium-upscale` sidecar (Aivium code)               | MIT                                                                                                             | (c) 2026 Aivium                                           | `sidecars/aivium-upscale` in the Aivium sources                          | yes (externalBin)                                                  |
| Real-ESRGAN-ncnn-vulkan (tiling scheme, options)     | MIT                                                                                                             | (c) 2021 Xintao Wang                                      | https://github.com/xinntao/Real-ESRGAN-ncnn-vulkan                       | derived code inside the sidecar                                    |
| waifu2x-ncnn-vulkan (CPU path)                       | MIT                                                                                                             | (c) 2018-2026 nihui                                       | https://github.com/nihui/waifu2x-ncnn-vulkan                             | derived code inside the sidecar                                    |
| ncnn (inference engine, in-house Vulkan loader)      | BSD-3-Clause (plus zlib-licensed parts, see upstream LICENSE.txt)                                               | (c) 2017 Tencent                                          | https://github.com/Tencent/ncnn                                          | statically linked; version pinned in `scripts/upscale-release.env` |
| stb_image, stb_image_write, stb_image_resize2        | public domain (Unlicense) or MIT                                                                                | Sean Barrett, Jeff Roberts, Jorge L Rodriguez et al.      | https://github.com/nothings/stb                                          | compiled into the sidecar                                          |
| glslang (bundled with ncnn)                          | BSD-3-Clause, BSD-2-Clause, MIT/Khronos, Apache-2.0; generated parser GPL-3.0-or-later WITH Bison-exception-2.2 | The Khronos Group Inc., 3Dlabs, LunarG, Google and others | https://github.com/nihui/glslang (fe88f421038e)                          | statically linked (inside ncnn)                                    |
| LLVM OpenMP runtime (libomp)                         | Apache-2.0 WITH LLVM-exception (older parts MIT OR NCSA)                                                        | LLVM Project contributors                                 | https://github.com/llvm/llvm-project/tree/main/openmp                    | macOS only, statically linked                                      |
| Microsoft Visual C++ OpenMP runtime (`vcomp140.dll`) | Microsoft redistributable license terms                                                                         | Microsoft                                                 | https://learn.microsoft.com/en-us/cpp/windows/latest-supported-vc-redist | Windows only, next to the sidecar                                  |

## Models

| Model                                      | License      | Copyright            | Source                                                                                                                                                                                                           | Bundled                                                                                                                                                                |
| ------------------------------------------ | ------------ | -------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `realesr-general-x4v3` ("Fast")            | BSD-3-Clause | (c) 2021 Xintao Wang | https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesr-general-x4v3.pth — converted to ncnn by Aivium; SHA256 of source and result in `sidecars/aivium-upscale/tools/convert-general-x4v3.md` | yes (resources `upscale-models/`)                                                                                                                                      |
| `realesrgan-x4plus` ("Quality")            | BSD-3-Clause | (c) 2021 Xintao Wang | https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesrgan-ncnn-vulkan-20220424-ubuntu.zip (`models/`)                                                                                         | no — downloaded on user action from `https://aivium.app/upscale-models/v1/`, mirror: the official archive above; SHA256 verified against `scripts/upscale-release.env` |
| `realesrgan-x4plus-anime` ("Illustration") | BSD-3-Clause | (c) 2021 Xintao Wang | same archive                                                                                                                                                                                                     | no — same download path                                                                                                                                                |

Downloaded models are stored in the Aivium app data directory under
`upscale-models/` and can be removed from Settings.

Not used, on purpose: `upscayl/upscayl-ncnn` (AGPL-3.0) and community models
from `upscayl/custom-models` (no license).

## License texts

Full texts of the MIT and BSD-3-Clause licenses with the copyright lines above
are kept in `sidecars/aivium-upscale/THIRD_PARTY_NOTICES.md`; they and the full
texts of ncnn, glslang, LLVM OpenMP and stb are reproduced in
`THIRD_PARTY_LICENSE_TEXTS.md` of the application's third-party notices.

This notice records the release engineering decision for Aivium's local
upscaling feature. It is not legal advice and does not relicense any
third-party component.