AI Cut models

ONNX weights used by AI Cut for video restoration, frame interpolation, smart crop / tracking, and source separation.

All files are served from this repository's master branch. The app verifies sha256 before installing. Download only the file you need.

Files

File Bytes SHA-256 License
rife/rife47_ensemble_True_scale_1_sim.onnx 21,459,153 f2479a88eae0ab33f697f3f481573ce5eacede039ade3088aee32ac091694972 MIT
real-esrgan-general-x4v3/realesr-general-x4v3.onnx 4,866,574 7964f4acc824de94b71240c6b81de424216a87b4bad8651d098c2e49ff57c2a0 BSD-3-Clause
real-esrgan-x4plus/RealESRGAN_x4plus.onnx 67,051,790 62f4ae8bd6313a4458ffaabcb5426fc2e4711648e89e967198528d9c523ede65 BSD-3-Clause
yolox-nano/yolox_nano.onnx 3,659,603 059e134ba0c599b025abc632530c1f730d624053277b214effc021940920f449 Apache-2.0
yunet/face_detection_yunet_2023mar.onnx 232,801 cea1903f83f897fd7bad52b7ec43ee97ea1c41a82c460bdc3444e9b1180e89aa MIT
uvr/UVR-MDX-NET-Voc_FT.onnx 66,762,795 e411182ce2c53541fefdcc99a8f46f2fe03978eb22038b9497c1d9d95a00fad4 UVR (see LICENSES/UVR-NOTICE.txt)
uvr/UVR_MDXNET_9482.onnx 29,704,738 9d78f8566fa8198065214ab628be1de966a500c57786695aa4b13e2b27a7727d UVR (see LICENSES/UVR-NOTICE.txt)

Model details

RIFE v4.7

  • Inputs: img0 [1,3,H,W] RGB float32, img1 [1,3,H,W] RGB float32, timestep [1] float32 in 0..1
  • Output: [1,3,H,W] RGB float32
  • Dynamic height/width.
  • License: MIT (LICENSES/RIFE-MIT.txt)
  • Upstream: https://github.com/hzwer/Practical-RIFE

Real-ESRGAN general-x4v3

  • Input: input [1,3,H,W] RGB float32 in 0..1
  • Output: [1,3,4H,4W] RGB float32
  • Dynamic height/width; 1.21M params; fast video tier.
  • License: BSD-3-Clause (LICENSES/Real-ESRGAN-BSD-3-Clause.txt)
  • Upstream: https://github.com/xinntao/Real-ESRGAN

Real-ESRGAN x4plus

  • Input: input [1,3,H,W] RGB float32 in 0..1
  • Output: [1,3,4H,4W] RGB float32
  • Dynamic height/width; 16.7M params; quality tier for short clips/stills.
  • License: BSD-3-Clause (LICENSES/Real-ESRGAN-BSD-3-Clause.txt)
  • Upstream: https://github.com/xinntao/Real-ESRGAN

YOLOX-nano

YuNet 2023mar

  • Input: input [1,3,640,640] BGR float32
  • Outputs: cls_8/16/32, obj_8/16/32, bbox_8/16/32, kps_8/16/32
  • Fixed 640x640; BGR channel order.
  • License: MIT (LICENSES/YuNet-MIT.txt)
  • Upstream: https://github.com/opencv/opencv_zoo

UVR MDX-Net

  • Input/output: float32[batch,4,dim_f,256] (see per-model metadata)
  • Do not strip ONNX metadata. sherpa-onnx reads model_type=UVR, stems, sample_rate, n_fft, center, win_length, hop_length, dim_t, dim_f, dim_c, window_type at load time.
  • Upstream: https://github.com/k2-fsa/sherpa-onnx/releases/tag/source-separation-models
  • The upstream model repository has no explicit license; see LICENSES/UVR-NOTICE.txt before commercial use.

Usage

import onnxruntime as ort
sess = ort.InferenceSession("rife/rife47_ensemble_True_scale_1_sim.onnx",
                            providers=["CPUExecutionProvider"])

For AI Cut, point the model catalog at these files and verify the SHA-256 above.

License

Each model retains its upstream license. See LICENSES/ and the table above.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support