AI super-resolution uses the AMD lightweight Real-ESRGAN 4x FP32 ONNX model.
Model: https://huggingface.co/amd/ryzenai-realesrgan
Revision: 8216bc22cd556dc073df761ca673fe70677dbf65
License declared by the model publisher: Apache-2.0 (see AI-Model-Apache-2.0.txt).
Model SHA256: 48D8425540F4D89650BF622ED4B059B705854A8CD108F41F8B339354C5042889
The model downloads on first use and runs locally with ONNX Runtime DirectML or CPU.
2x is AI 4x followed by reduction. 8x uses AI 4x/reduction to 2x, then a second AI 4x pass.
Real-ESRGAN research: https://arxiv.org/abs/2107.10833
Our application implements its own inference tiling, processing workflow and interface.
Movavi code, models, trademarks and icon assets are not included.
ONNX Runtime and DirectML license and third-party notices are supplied alongside this file.