jd7h/propainter ✓🖼️❓🔢 → 🖼️

▶️ 49.8K runs 📅 Oct 2023 ⚙️ Cog 0.8.6 🔗 GitHub 📄 Paper ⚖️ License
video-inpainting video-outpainting

About

Object removal, video completion and video outpainting

Example Output

Output

Performance Metrics

11.52s Prediction Time
11.51s Total Time
All Input Parameters
{
  "fp16": true,
  "mask": "https://replicate.delivery/pbxt/JgGNe7IjsavH2is00AVgbEwWroEtseid6f3HZIHI7MJH9Num/bmx-trees-mask4.mp4",
  "mode": "video_inpainting",
  "video": "https://replicate.delivery/pbxt/JgGNdl0r9V14YxgJ4JlsCz71rSU9y8NJD9RDDdJo2Ua3r8C7/bmx-trees2.mp4",
  "width": -1,
  "height": -1,
  "scale_h": 1,
  "scale_w": 1,
  "save_fps": 24,
  "raft_iter": 20,
  "ref_stride": 10,
  "resize_ratio": 1,
  "mask_dilation": 4,
  "neighbor_length": 10,
  "subvideo_length": 80,
  "return_input_video": true
}
Input Parameters
fp16 Type: booleanDefault: false
Use fp16 (half precision) during inference. Default: fp32 (single precision).
mask Type: string
Mask for video inpainting. Can be a static image (jpg, png) or a video (avi, mp4).
mode Default: video_inpainting
Modes: video inpainting / video outpainting. If you want to do video inpainting, you need a mask. For video outpainting, you need to set scale_h and scale_w, and mask is ignored.
video (required) Type: string
Input video
width Type: integerDefault: -1
Width of the processing video.
height Type: integerDefault: -1
Height of the processing video.
scale_h Type: numberDefault: 1
Outpainting scale of height for video_outpainting mode.
scale_w Type: numberDefault: 1
Outpainting scale of width for video_outpainting mode.
save_fps Type: integerDefault: 24
Frames per second.
raft_iter Type: integerDefault: 20
Iterations for RAFT inference.
ref_stride Type: integerDefault: 10
Stride of global reference frames.
resize_ratio Type: numberDefault: 1
Resize scale for processing video.
mask_dilation Type: integerDefault: 4
Mask dilation for video and flow masking.
neighbor_length Type: integerDefault: 10
Length of local neighboring frames.
subvideo_length Type: integerDefault: 80
Length of sub-video for long video inference.
return_input_video Type: booleanDefault: false
Return the input video in the output.
Output Schema

Output

Type: arrayItems Type: stringItems Format: uri

Example Execution Logs
Interpreting mask as a video...
Processing: tmppyc3cmpbbmx-trees2 [80 frames]...
  0%|          | 0/16 [00:00<?, ?it/s]
  6%|▋         | 1/16 [00:00<00:02,  6.11it/s]
 12%|█▎        | 2/16 [00:00<00:02,  5.13it/s]
 19%|█▉        | 3/16 [00:00<00:02,  4.86it/s]
 25%|██▌       | 4/16 [00:00<00:02,  4.89it/s]
 31%|███▏      | 5/16 [00:01<00:02,  4.86it/s]
 38%|███▊      | 6/16 [00:01<00:02,  4.89it/s]
 44%|████▍     | 7/16 [00:01<00:01,  4.84it/s]
 50%|█████     | 8/16 [00:01<00:01,  4.86it/s]
 56%|█████▋    | 9/16 [00:01<00:01,  4.76it/s]
 62%|██████▎   | 10/16 [00:02<00:01,  4.80it/s]
 69%|██████▉   | 11/16 [00:02<00:01,  4.81it/s]
 75%|███████▌  | 12/16 [00:02<00:00,  4.86it/s]
 81%|████████▏ | 13/16 [00:02<00:00,  4.85it/s]
 88%|████████▊ | 14/16 [00:02<00:00,  4.90it/s]
 94%|█████████▍| 15/16 [00:03<00:00,  4.91it/s]
100%|██████████| 16/16 [00:03<00:00,  5.02it/s]
100%|██████████| 16/16 [00:03<00:00,  4.91it/s]
All results are saved in results/tmppyc3cmpbbmx-trees2
Version Details
Version ID
e5ea7ae04e97c96a0e14c70d8e4cb899abdf326a377c01f1c10966ccd6c6bae4
Version Created
October 11, 2023
Run on Replicate →