Inpaint Anything can inpaint anything in images, videos and 3D scenes!
- Authors: Tao Yu, Runseng Feng, Ruoyu Feng, Jinming Liu, Xin Jin, Wenjun Zeng and Zhibo Chen.
- Institutes: University of Science and Technology of China; Eastern Institute for Advanced Study.
- [Paper] [Website] [Hugging Face Homepage]
TL; DR: Users can select any object in an image by clicking on it. With powerful vision models, e.g., SAM, LaMa and Stable Diffusion (SD), Inpaint Anything is able to remove the object smoothly (i.e., Remove Anything). Further, prompted by user input text, Inpaint Anything can fill the object with any desired content (i.e., Fill Anything) or replace the background of it arbitrarily (i.e., Replace Anything).
🚧 New: main_2026 branch — modernized stack + robotics support (beta)
The main_2026
branch brings Inpaint Anything up to date with the 2026 model landscape, and adds
a new direction: data engineering for robotics.
main (this branch) |
main_2026 |
|
|---|---|---|
| Segmentation | SAM 1 | SAM 3 — plus open-vocabulary text prompts |
| Video / 3D tracking | OSTrack | SAM 3 video predictor — one less model and checkpoint |
| Video inpainting | STTN | ProPainter |
| Text-guided fill / replace | SD 2 (no longer downloadable) | SDXL, optional FLUX.1-Fill |
| Robotics | — | remove_hands.py — batch hand removal for Human-to-Robot pipelines |
Two things you can do there that you cannot do here:
- Name the object instead of clicking it.
--text_select "dog"finds every match, which also means the pipelines can run unattended over a whole dataset. - Prepare egocentric data for robot learning.
remove_hands.pyerases human hands from egocentric video and exports the masks — the hand removal and inpainting stage of Human-to-Robot synthesis pipelines such as Qwen-RobotManip and EgoEngine. On EgoMimic footage it matches human annotation at IoU 0.96, and it reconstructs the background rather than blacking the arm out.
git checkout main_2026
# then follow the Quick start in that branch's README
⚠️ main_2026is in beta. It needs Python ≥ 3.12, PyTorch ≥ 2.7 and CUDA ≥ 12.6 (SAM 3's floor), and the NeRF-based 3D path has not been end-to-end verified on that stack yet. Every legacy backend (SAM 1 / MobileSAM, OSTrack, STTN) is still selectable by flag, so you can fall back per stage.🤝 Contributions very welcome — especially on the robotics side. Issues and PRs against
main_2026are appreciated: more egocentric datasets, action retargeting, robot rendering and compositing, or newer inpainting backends. Please open an issue if you hit anything.
[2026/7/28] 🔥NEW main_2026 branch (beta): upgraded to SAM 3 with text prompts, ProPainter for video, SDXL/FLUX for text-guided editing, and robotics support via remove_hands.py. OSTrack is no longer needed. Contributions welcome!
[2023/9/15] Remove Anything 3D code is available!
[2023/4/30] Remove Anything Video available! You can remove any object from a video!
[2023/4/24] Local web UI supported! You can run the demo website locally!
[2023/4/22] Website available! You can experience Inpaint Anything through the interface!
[2023/4/22] Remove Anything 3D available! You can remove any 3D object from a 3D scene!
[2023/4/13] Technical report on arXiv available!
- Remove Anything
- Fill Anything
- Replace Anything
- Remove Anything 3D (🔥NEW)
- Fill Anything 3D
- Replace Anything 3D
- Remove Anything Video (🔥NEW)
- Fill Anything Video
- Replace Anything Video
- Any aspect ratio supported
- 2K resolution supported
- Technical report on arXiv available (🔥NEW)
- Website available (🔥NEW)
- Local web UI available (🔥NEW)
- Multiple modalities (i.e., image, video and 3D scene) supported (🔥NEW)
Click on an object in the image, and Inpainting Anything will remove it instantly!
- Click on an object;
- Segment Anything Model (SAM) segments the object out;
- Inpainting models (e.g., LaMa) fill the "hole".
Requires python>=3.8
python -m pip install torch torchvision torchaudio
python -m pip install -e segment_anything
python -m pip install -r lama/requirements.txt In Windows, we recommend you to first install miniconda and
open Anaconda Powershell Prompt (miniconda3) as administrator.
Then pip install ./lama_requirements_windows.txt instead of
./lama/requirements.txt.
Download the model checkpoints provided in Segment Anything and LaMa (e.g., sam_vit_h_4b8939.pth and big-lama), and put them into ./pretrained_models. For simplicity, you can also go here, directly download pretrained_models, put the directory into ./ and get ./pretrained_models.
For MobileSAM, the sam_model_type should use "vit_t", and the sam_ckpt should use "./weights/mobile_sam.pt". For the MobileSAM project, please refer to MobileSAM
bash script/remove_anything.sh
Specify an image and a point, and Remove Anything will remove the object at the point.
python remove_anything.py \
--input_img ./example/remove-anything/dog.jpg \
--coords_type key_in \
--point_coords 200 450 \
--point_labels 1 \
--dilate_kernel_size 15 \
--output_dir ./results \
--sam_model_type "vit_h" \
--sam_ckpt ./pretrained_models/sam_vit_h_4b8939.pth \
--lama_config ./lama/configs/prediction/default.yaml \
--lama_ckpt ./pretrained_models/big-lamaYou can change --coords_type key_in to --coords_type click if your machine has a display device. If click is set, after running the above command, the image will be displayed. (1) Use left-click to record the coordinates of the click. It supports modifying points, and only last point coordinates are recorded. (2) Use right-click to finish the selection.
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
Text prompt: "a teddy bear on a bench"
Click on an object, type in what you want to fill, and Inpaint Anything will fill it!
- Click on an object;
- SAM segments the object out;
- Input a text prompt;
- Text-prompt-guided inpainting models (e.g., Stable Diffusion) fill the "hole" according to the text.
Requires python>=3.8
python -m pip install torch torchvision torchaudio
python -m pip install -e segment_anything
python -m pip install diffusers transformers accelerate scipy safetensorsDownload the model checkpoints provided in Segment Anything (e.g., sam_vit_h_4b8939.pth) and put them into ./pretrained_models. For simplicity, you can also go here, directly download pretrained_models, put the directory into ./ and get ./pretrained_models.
For MobileSAM, the sam_model_type should use "vit_t", and the sam_ckpt should use "./weights/mobile_sam.pt". For the MobileSAM project, please refer to MobileSAM
bash script/fill_anything.sh
Specify an image, a point and text prompt, and run:
python fill_anything.py \
--input_img ./example/fill-anything/sample1.png \
--coords_type key_in \
--point_coords 750 500 \
--point_labels 1 \
--text_prompt "a teddy bear on a bench" \
--dilate_kernel_size 50 \
--output_dir ./results \
--sam_model_type "vit_h" \
--sam_ckpt ./pretrained_models/sam_vit_h_4b8939.pth |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
Text prompt: "a man in office"
Click on an object, type in what background you want to replace, and Inpaint Anything will replace it!
- Click on an object;
- SAM segments the object out;
- Input a text prompt;
- Text-prompt-guided inpainting models (e.g., Stable Diffusion) replace the background according to the text.
Requires python>=3.8
python -m pip install torch torchvision torchaudio
python -m pip install -e segment_anything
python -m pip install diffusers transformers accelerate scipy safetensorsDownload the model checkpoints provided in Segment Anything (e.g. sam_vit_h_4b8939.pth) and put them into ./pretrained_models. For simplicity, you can also go here, directly download pretrained_models, put the directory into ./ and get ./pretrained_models.
For MobileSAM, the sam_model_type should use "vit_t", and the sam_ckpt should use "./weights/mobile_sam.pt". For the MobileSAM project, please refer to MobileSAM
bash script/replace_anything.sh
Specify an image, a point and text prompt, and run:
python replace_anything.py \
--input_img ./example/replace-anything/dog.png \
--coords_type key_in \
--point_coords 750 500 \
--point_labels 1 \
--text_prompt "sit on the swing" \
--output_dir ./results \
--sam_model_type "vit_h" \
--sam_ckpt ./pretrained_models/sam_vit_h_4b8939.pth |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
With a single click on an object in the first view of source views, Remove Anything 3D can remove the object from the whole scene!
- Click on an object in the first view of source views;
- SAM segments the object out (with three possible masks);
- Select one mask;
- A tracking model such as OSTrack is ultilized to track the object in these views;
- SAM segments the object out in each source view according to tracking results;
- An inpainting model such as LaMa is ultilized to inpaint the object in each source view.
- A novel view synthesizing model such as NeRF is ultilized to synthesize novel views of the scene without the object.
Requires python>=3.8
python -m pip install torch torchvision torchaudio
python -m pip install -e segment_anything
python -m pip install -r lama/requirements.txt
python -m pip install jpeg4py lmdbDownload the model checkpoints provided in Segment Anything and LaMa (e.g., sam_vit_h_4b8939.pth), and put them into ./pretrained_models. Further, download OSTrack pretrained model from here (e.g., vitb_384_mae_ce_32x4_ep300.pth) and put it into ./pytracking/pretrain. In addition, download [nerf_llff_data] (e.g, horns), and put them into ./example/3d. For simplicity, you can also go here, directly download pretrained_models, put the directory into ./ and get ./pretrained_models. Additionally, download pretrain, put the directory into ./pytracking and get ./pytracking/pretrain.
For MobileSAM, the sam_model_type should use "vit_t", and the sam_ckpt should use "./weights/mobile_sam.pt". For the MobileSAM project, please refer to MobileSAM
bash script/remove_anything_3d.sh
Specify a 3d scene, a point, scene config and mask index (indicating using which mask result of the first view), and Remove Anything 3D will remove the object from the whole scene.
python remove_anything_3d.py \
--input_dir ./example/3d/horns \
--coords_type key_in \
--point_coords 830 405 \
--point_labels 1 \
--dilate_kernel_size 15 \
--output_dir ./results \
--sam_model_type "vit_h" \
--sam_ckpt ./pretrained_models/sam_vit_h_4b8939.pth \
--lama_config ./lama/configs/prediction/default.yaml \
--lama_ckpt ./pretrained_models/big-lama \
--tracker_ckpt vitb_384_mae_ce_32x4_ep300 \
--mask_idx 1 \
--config ./nerf/configs/horns.txt \
--expname hornsThe --mask_idx is usually set to 1, which typically is the most confident mask result of the first frame. If the object is not segmented out well, you can try other masks (0 or 2).
 |
 |
 |
With a single click on an object in the first video frame, Remove Anything Video can remove the object from the whole video!
- Click on an object in the first frame of a video;
- SAM segments the object out (with three possible masks);
- Select one mask;
- A tracking model such as OSTrack is ultilized to track the object in the video;
- SAM segments the object out in each frame according to tracking results;
- A video inpainting model such as STTN is ultilized to inpaint the object in each frame.
Requires python>=3.8
python -m pip install torch torchvision torchaudio
python -m pip install -e segment_anything
python -m pip install -r lama/requirements.txt
python -m pip install jpeg4py lmdbDownload the model checkpoints provided in Segment Anything and STTN (e.g., sam_vit_h_4b8939.pth and sttn.pth), and put them into ./pretrained_models. Further, download OSTrack pretrained model from here (e.g., vitb_384_mae_ce_32x4_ep300.pth) and put it into ./pytracking/pretrain. For simplicity, you can also go here, directly download pretrained_models, put the directory into ./ and get ./pretrained_models. Additionally, download pretrain, put the directory into ./pytracking and get ./pytracking/pretrain.
For MobileSAM, the sam_model_type should use "vit_t", and the sam_ckpt should use "./weights/mobile_sam.pt". For the MobileSAM project, please refer to MobileSAM
bash script/remove_anything_video.sh
Specify a video, a point, video FPS and mask index (indicating using which mask result of the first frame), and Remove Anything Video will remove the object from the whole video.
python remove_anything_video.py \
--input_video ./example/video/paragliding/original_video.mp4 \
--coords_type key_in \
--point_coords 652 162 \
--point_labels 1 \
--dilate_kernel_size 15 \
--output_dir ./results \
--sam_model_type "vit_h" \
--sam_ckpt ./pretrained_models/sam_vit_h_4b8939.pth \
--lama_config lama/configs/prediction/default.yaml \
--lama_ckpt ./pretrained_models/big-lama \
--tracker_ckpt vitb_384_mae_ce_32x4_ep300 \
--vi_ckpt ./pretrained_models/sttn.pth \
--mask_idx 2 \
--fps 25The --mask_idx is usually set to 2, which typically is the most confident mask result of the first frame. If the object is not segmented out well, you can try other masks (0 or 1).
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
If you find this work useful for your research, please cite us:
@article{yu2023inpaint,
title={Inpaint Anything: Segment Anything Meets Image Inpainting},
author={Yu, Tao and Feng, Runseng and Feng, Ruoyu and Liu, Jinming and Jin, Xin and Zeng, Wenjun and Chen, Zhibo},
journal={arXiv preprint arXiv:2304.06790},
year={2023}
}
