Segment Anything for Stable Diffusion WebUI
This extension aim for connecting AUTOMATIC1111 Stable Diffusion WebUI and Mikubill ControlNet Extension with segment anything and GroundingDINO to enhance Stable Diffusion/ControlNet inpainting, enhance ControlNet semantic segmentation, automate image matting and create LoRA/LyCORIS training set.
2023/04/10
: v1.0.0 SAM extension released! You can click on the image to generate segmentation masks.2023/04/12
: v1.0.1 Mask expansion and API support released by @jordan-barrett-jm! You can expand masks to overcome edge problems of SAM.2023/04/15
: v1.1.0 GroundingDINO support released! You can enter text prompts to generate bounding boxes and segmentation masks.2023/04/18
: v1.2.0 ControlNet V1.1 inpainting support released! You can copy SAM generated masks to ControlNet to do inpainting. Note that you must update ControlNet extension to use it. ControlNet inpainting has far better performance compared to general-purposed models, and you do not need to download inpainting-specific models anymore.2023/04/24
: v1.3.0 Automatic segmentation support released! Functionalities with * require you to have ControlNet extension installed. This update includes support for
2023/04/29
: v1.4.0 API has been completely refactored. You can access all features for single image process through API. API documentation has been moved to wiki.2023/05/22
: v1.4.1 EditAnything is ready to use! You can generate random segmentation and copy the output to EditAnything ControlNet.2023/05/29
: v1.4.2 You may now do SAM inference on CPU by checking "Use CPU for SAM". This is for some MAC users who are not able to do SAM inference on GPU. I discourage other users from using this feature because it is significantly slower than CUDA.2023/06/01
: v1.5.0 You may now choose to use local GroundingDINO to bypass C++ problem. See FAQ-1 for more detail.2023/06/04
: v1.5.1 Upload Mask to ControlNet Inpainting
comes back in response to ControlNet inpaint improvement. You should see a new tab beside AutoSAM
after updating the extension. This feature will again be removed once ControlNet extension has its own uploading feature.2023/06/13
: v1.6.0 SAM-HQ supported by @SpenserCai and me. This is an "upgraded" SAM, created by researchers at ETH Zurich & HKUST. However, I cannot guarantee which one is better and you should make your own choice based on your own experiments. Go to Installation to get the link to the models.2023/06/29
: v1.6.1 MobileSAM supported. This is a tiny version of SAM, created by researchers at Kyung Hee University. Visit here for more information.2023/08/31
: v1.6.2 Support WebUI v1.6.0, Gradio v3.41.2Note that support for some other variations of SAM, such as Matting-Anything and FastSAM are still on the way. Support for these models, unlike MobileSAM, are non-trivial, especially FastSAM, which utilize a completely different pipeline, ultralytics/YOLO. Introducing these new works to the current codebase will make the original ugly-enough codebase more ugly. They will be supported once I finish a major refactor of the current codebase.
Thanks for suggestions from github issues, reddit and bilibili to make this extension better.
There are already at least two great tutorials on how to use this extension. Check out this video (Chinese) from @ThisisGameAIResearch and this video (Chinese) from @OedoSoldier. You can also check out my demo.
You should know the following before submitting an issue.
Due to the overwhelming complaints about GroundingDINO installation and the lack of substitution of similar high-performance text-to-bounding-box library, I decide to modify the source code of GroundingDINO and push to this repository. Starting from v1.5.0, you can choose to use local GroundingDINO by checking Use local groundingdino to bypass C++ problem
on Settings/Segment Anything
. This change should solve all problems about ninja, pycocotools, _C and any other problems related to C++/CUDA compilation.
If you did not modify the setting described above, This script will firstly try to install GroundingDINO and check if your environment has successfully built the C++ dynamic library (the annoying _C
). If so, this script will use the official implementation of GroundingDINO. This is to show respect to the authors of GroundingDINO. If the script failed to install GroundingDINO, it will use local GroundingDINO instead.
If you'd still like to resolve the install problem of GroundingDINO, I observe some common problems for Windows users:
If you are still unable to install GroundingDINO on Windows AND you cannot resolve this problem AFTER searching for issues inside here here and here, You may refer to #98 and watch the videos there. Note that I develop on linux, so I cannot guarantee that any video tutorials may or may not work.
If you
The problem is most likely due to some other extensions which might also change the position inside the extension list to control ControlNet. The easiest solution is here. This change will precede SAM extension to be before ControlNet, bypassing the internal preceding code, and will not prevent you from receiving any updates from me. I am not planning to refactor my code to bypass this problem. I did not expect to control ControlNet when I created this extension, but ControlNet indeed grow much faster than my expectation.
This extension has almost moved into maintenance phase. Although I don't think there will be huge updates in the foreseeable future, Mikubill ControlNet Extension is still fast developing, and I'm looking forward to more opportunities to connect my extension to ControlNet. Despite of this, I will continue to deal with issues, and monitor new research works to see if they are worth supporting. I welcome any community contribution and any feature requests.
You must use gradio>=3.23.0 and WebUI>=22bcc7be
to use this extension. A1111 WebUI is stable, and some integrated package authors have also updated their packages (for example, if you are using the package from @Akegarasu, i.e. 秋叶整合包, it has already been updated according to this video). Also, supporting different versions of WebUI will be a huge time commitment, during which I can create many more features. Please update your WebUI and it is safe to use. I'm not planning to support some old commits of WebUI, such as a9fed7c3
.
It is impossible to support the following features, at least at this moment, due to gradio/A1111 limitations. I will closely monitor gradio/A1111 update to see if it becomes possible to support them:
Inpaint-Anything and EditAnything and A LOT of other popular SAM extensions have been supported. For Inpaint-Anything, you may check this issue for how to use. For EditAnything, please check how to use. I am always open to support any other interesting applications, submit a feature request if you find another interesting one.
If you have a job opportunity and think I am a good fit, please feel free to send me an email.
If you want to sponsor me, please go to sponsor section and scan the corresponding QR code.
Download this extension to ${sd-webui}/extensions
via whatever way you like (git clone or install from UI)
Choose one or more of the models below and put them to ${sd-webui}/models/sam
or ${sd-webui-segment-anything}/models/sam
(Choose one, not both. Remove the former folder if you choose to use the latter.). Do not change model name, otherwise this extension may fail due to a bug inside segment anything.
We support several variations of segmentation models:
SAM from Meta AI.
I myself tested vit_h on NVIDIA 3090 Ti which is good. If you encounter VRAM problem, you should switch to smaller models.
SAM-HQ from SysCV.
MobileSAM from Kyung Hee University.
We plan to (NOT supported yet) support some other variations of segmentation models after a major refactor of the codebase:
Matting-Anything from SHI-Labs. This is a post-processing model for any variation of SAM. Put the model under ${sd-webui-segment-anything}/models/sam
FastSAM from CASIA-IVA-Lab. This is a YOLO variation of SAM.
GroundingDINO packages, GroundingDINO models and ControlNet annotator models will be automatically installed the first time you use them.
If your network does not allow you to access huggingface via the terminal, download GroundingDINO models from huggingface and put them under ${sd-webui-segment-anything}/models/grounding-dino
. Please note that GroundingDINO still need to access huggingface to download bert vocabularies. There is no alternative at this time. Read here to find a way to resolve this problem. I will try to find an alternative in the near future.
GroundingDINO has been supported in this extension. It has the following functionalities:
.
. SAM can convert these bounding boxes to masksBatch Process
tab to do image matting and get LoRA/LyCORIS training setHowever, there are some existing problems with GroundingDINO:
${sd-webui-segment-anything}/models/grounding-dino
._C
problem, it's most probably because you did not install CUDA Toolkit. Follow steps decribed here. DO NOT skip steps. Otherwise, please go to Grounded-SAM Issue Page and submit an issue there. Despite of this, you can still use this extension for point prompts->segmentation masks even if you cannot install GroundingDINO, don't worry.For more detail, check How to Use and Demo.
Automatic Segmentation has been supported in this extension. It has the following functionalities:
However, there are some existing problems with AutoSAM:
sd-webui-controlnet
).seg_ufade20k
and SAM. You may only observe some slight improvement if you combine one of the Oneformer
preprocessors (seg_ofade20k
&seg_ofcoco
). This is because Oneformer is already very strong, compared to Uniformer, for semantic segmentation. SAM can only improve some details of semantic segmentation instead of showing some categories semantic models cannot show, because SAM is NOT a semantic-recognizable model.If you have previously enabled other copies while using this extension, you may want to click Uncheck all copies
at the bottom of this extension UI, to prevent other copies from affecting your current page.
Enable GroundingDINO
, select GroundingDINO model you want, write text prompt (separate different categories with .
) and pick a box threshold (I highly recommend the default setting. High threshold may result in no bounding box). You must write text prompt if you do not wish to use point prompts.Generate bounding box
. You must write text prompt to preview bounding box. After you see the boxes with number marked on the top-left corner, uncheck all the boxes you do not want. If you uncheck all boxes, you will have to add point prompts to generate masks.Preview Segmentation
button. Due to the limitation of SAM, if there are multiple bounding boxes, your point prompts will not take effect when generating masks.Expand Mask
and specify the amount, then click Update Mask
.Allow other script to control this extension
(MUST) on your ControlNet settings.ControlNet inpaint not masked
to invert mask colors and inpaint regions outside of the mask.Enable
, preprocessor choose inpaint_global_harmonious
, model choose control_v11p_sd15_inpaint [ebff9138]
. There is no need to upload image to the ControlNet inpainting panel.Generate
.Copy to Inpaint Upload & ControlNet Inpainting
. There is no need to select ControlNet index.Enable
, preprocessor choose inpaint_global_harmonious
, model choose control_v11p_sd15_inpaint [ebff9138]
. There is no need to upload image to the ControlNet inpainting panel.Switch to Inpaint Upload
button. There is no need to upload another image or mask, just leave them blank. Write your prompts, configurate A1111 panel and click Generate
.Output per image
to configurate the number of masks per bounding box. I highly recommend 3, since some masks might be wierd.Start batch process
and wait. If you see "Done" below this button, you are all set.seg_ufade20k
, seg_ofade20k
and seg_ofcoco
are from ControlNet annotators. I highly recommend one of seg_ofade20k
and seg_ofcoco
because their performance are far better than seg_ufade20k
. They are all compatible with control_v11p_sd15_seg
. Optionally enable pixel-perfect to automatically pick the best preprocessor resolution. Configure your target width and height on txt2img/img2img default panel before preview if you wish to enable pixel perfect. Otherwise you need to manually set a preprocessor resolution.random
is for EditAnything. There is no need to set preprocessor resolution for random preprocessor since it does not contain semantic segmentation, but you need to pick an image from the AutoSeg output gallery to copy to ControlNet. 1 represents random colorization of different mask regions which is reserved for future ControlNet, 2 represents fixed colorization which can be EditAnything ControlNet control image.Copy to ControlNet Segmentation
and select the correct ControlNet index where you are using ControlNet segmentation models if you wish to use Multi-ControlNet.Enable
, preprocessor choose none
, model choose control_v11p_sd15_seg [e1f51eb9]
. There is no need to upload image to the ControlNet segmentation panel.Generate
.random
preprocessor.control_v11p_sd15_seg [e1f51eb9]
${destination}/{image_filename}
directory.+
. Visit here for ade20k and here for coco to get category->id map. Note that coco jumps some numbers, so the actual ID is line_number - 21. For example, if you want bed+person, your input should be 7+12 for ade20k and 59+0 for coco.Point prompts demo (also so-called Remove/Fill Anything)
GroundingDINO demo
Batch process demo
Input Image | Output Image | Output Mask | Output Blend |
---|---|---|---|
Semantic segmentation demo
Mask by Category demo (also so-called Replace Anything)
Mask by Category batch demo
Input Image | Output Image | Output Mask | Output Blend |
---|---|---|---|
Disclaimer: I have not thoroughly tested this extension, so there might be bugs. Bear with me while I'm fixing them :)
If you encounter a bug, please submit an issue. Please at least provide your WebUI version, your extension version, your browser version, errors on your browser console log if there is any, error on your terminal log if there is any, to save both of our time.
I welcome any contribution. Please submit a pull request if you want to contribute
You can sponsor me via WeChat, AliPay or PayPal.
AliPay | PayPal | |
---|---|---|