Home Creators Posts Import Register Favorites Logout

Downloads

Content

BiRefNet HR removes backgrounds at high resolution in single or batch mode, with BF16, FP16 or FP32 precision and a No Resize mode that keeps native size.

Supports the High Resolution model and all other BiRefNet versions, preset save and load, and replacing transparent pixels with white or black. Includes 1-click installers for Windows, RunPod and Massed Compute and a free Kaggle account notebook, updated to Torch 2.8 for all GPUs.

Amazing Gradio APP for the very best open source background remover model BiRefNet HR Version - 1-Click to install on Windows, RunPod, Kaggle

Please visit Patreon exclusive posts index to find all of our Patreon scripts, posts, apps and more.

Join discord to get help, chat, discuss and also tell me your discord username to get your special rank : SECourses Discord

Please also Star, Watch and Fork our Stable Diffusion & Generative AI  GitHub repository and join our Reddit subreddit and follow me on LinkedIn (my real profile)

=======

Latest installer zip file: BiRefNet_HR_v10.zip

  • GitHub official repo : https://github.com/ZhengPeng7/BiRefNet

  • We have upgraded our app to work better and now supports BiRefNet HR - High Resolution version and all other versions as well

  • Moreover newer version has half precision feature so now it works even with lower VRAM

    • VRAM usage 100% depends on the processed image resolution

  • Check all the screenshots to see how it works and we have full batch processing feature as well

  • How to install and use shown in this tutorial : https://youtu.be/MLNQkwW_Q8M?si=8FoMpuIqU68DurJM&t=248

25 November 2025 Update v10

  • How to install and use shown in this tutorial : https://youtu.be/MLNQkwW_Q8M?si=8FoMpuIqU68DurJM&t=248

  • All installers updated to latest Torch 2.8 for Windows, RunPod and Massed Compute with latest pre-compiled Flash Attention, Sage Attention and xFormers for All GPUs such as RTX 3000 series, 4000 series, 5000 series, H100, B200, RTX PRO 6000 etc

  • Also we have updated latest changes from the official repo like BF16, original image resolution processing and more improvements

    • You can see comparison here : https://imgsli.com/NDI5MzY0/0/1

      • BF16 mixed precision is now the default; the app lets you switch between FP32, FP16, and BF16, with inference handled via autocast rather than converting model weights.

      • “Original (No Resize)” mode truly preserves native resolution while internally padding to multiple-of-32 dimensions so the decoder never crashes.

      • Resolution presets (Original, 1024, 2048, 4096, Custom) plus a dedicated custom textbox make resizing explicit and easier to control in both single and batch tabs.

      • Output files can include automatic model/resolution metadata (results/<model>-reso_<size>/...), keeping experiments organized without manual folder management.

      • The Gradio UI was refreshed with the new options (precision dropdown, resolution preset/custom toggle, metadata checkbox) and updated copy pointing to the Patreon release.

  • This is currently literally the very fast and highest quality background remover model

  • New preset save and load feature added

    • It will auto load your last used preset when you start app next time

  • New replace transparent pixels with white or black feature added

  • I recommend a fresh install

Windows Requirements

  • Python 3.10, FFmpeg, CUDA 12.9, cuDNN 9.12 , C++ tools and Git

  • If it doesn't work make sure to below tutorial and install everything exactly as shown in this below tutorial

  • https://youtu.be/DrhUHnYfwC0

How To Install and Use:

Windows:

  • First watch above requirements tutorial and follow every step

    • Extract zip archive into a drive like c:/BiRefNet_hr

    • Longer directory path causes issues and don't use space characters in folder names

    • Double click and install with Windows_Install.bat and that is it

    • Then run with Windows_Start_App.bat to start BiRefNet HR SECourses APP

    • And follow above and below attached screenshot

Massed Compute (Recommend Cloud) :

RunPod (Cloud):

Kaggle (Free Cloud - Not Recommend):

Gradio Live Share now working so we don't need Ngrok Tunneling anymore

  

 

Related posts

All posts in this topic: AI Image Generation and Editing Apps · New to SECourses? Start Here: Best and Most Popular SECourses Posts

Files

Comments

cool1

So can this be used instead of rotoscoping for removing a background from a video of a person (eg. if someone is singing in a room in a 1080wx1920h 30 fps video of about 3 mins) can this remove the background of the video? edit: I've tested it with a few images and so far it seems really good at it. I'm exporting a video to test with it now.

GenerativeJoy

Thank you for your work and for making the script so soon

Markus

What does ist do?

cool1

removes the background in images.

SECourses

you can remove it on exported video frames. should work fairly well

SECourses

You are welcome thanks for support

SECourses

It removes background and it is the best one

cool1

Thanks. It's still exporting them then I'll try that. I tried it with 1 frame from the video and it seems mostly okay except part of the microphone stand got partially transparent (probably because it's a similar colour to the wall behind). It would be good if we could also change things like the colour of someone's shirt without affecting other things in the video but I assume that wouldn't be possible with this one. I was also wondering if for video it would be consistent enough. Hopefully it will.

SECourses

ye that is another task. i am following maybe it will arrive too

cool1

It's converting the images from my video. The conversions are mostly roughly okay (sometimes it can't tell which bits are part of the background (eg. maybe if there's not enough light in the background or enough brightness difference) but generally it's okay. One thing is, it puts each new converted image as an image into the "output image" listbox when doing batch processing. If I'm converting about 5000 source images (or more) isn't adding them all to that image list slowing the app down (and maybe using more memory for each one - though I have plenty of RAM but less VRAM)? Maybe it might go a bit faster if only the filename was added to the listbox in batch mode or it only showed the currently converted filename or image (or the last few) when in batch mode. That way it shouldn't get slower with each new image when converting hundreds or thousands of images (eg. video frames) in a batch

SECourses

I added a checkbox to batch processing. run Windows_Update_App.bat and restart the app you will see it - Display Images During Batch Processing

Dazzastrous

the other version gives better results for me

SECourses

we have both versions but this works better for me

George Gostyshev

Nice addition to RMBG2. Thanks! Have a question\request. May be there's a way with this technology or any other to do good quality segmentation for complex scenes? Could be nice app to have)

George Gostyshev

Great! Didn't see it. Thanks will check it out now

cool1

For changing the colour of the shirt, I used your Segmenter/Sam2 one after I saw that today, by telling it to mask "shirt" and then recolouring just the masked section in After Effects and that worked okayish (and also used that for the person. I tried BiRef for the person too but segmenter was faster). It missed some parts sometimes but I didn't try different values for the 2 settings. Maybe if I had that could have helped. But it's a lot faster than trying to do it all manually. Maybe some parts could be fixed manually too. So there's probably some types of shots that it could do it all accurately I think. Ideally there could be a way for the segmenter to feather the edge a bit (but maybe that can be done in other things like After Effects).

BecauseReasons

I have Python 3.11, how come we're still using 3.10? Install does not work.

SECourses

because 3.10 is most commonly supported python. for example kohya still doesnt work with 3.11 afaik. have 3.10 in your system my installer will auto use it. it doesnt have to be default. the tutorial video shows it. when kaggle and google colab moves to python 3.11 i think i will move as well

Realtoonai

shoud I remove v1 before install this v2 ?

Mig Test

Feature request: Could you please add an option save without an alpha channel? e.g. set pixels that would have been 'clear' to 0,0,0 for black or 255,255,255 for white?

SECourses

just added with v10 and also i added preset save and load