Demucs v3 Has been fully added to the UVR GUI!
Anjok07 opened this issue · comments
This isn't an issue, just an announcement.
Update: 01/01/23
Demucs v3/v4 has been fully implemented into the UVR GUI. You can separate all 4 stems and add new models easily. I also included a 2 stem vocal/instrumental Demucs v3 models we trained ourselves with the UVR dataset! In addition to this, I also included full backward compatibility with Demucs v1 and v2 models in the latest update. To add more Demucs models, you can download them straight from the application via the "Download Center" tab within the "Settings" menu.
Stand-alone bundles are available for Windows and MacOS.
You can find them here - https://github.com/Anjok07/ultimatevocalremovergui#installation
Hello. This is an amazing program and I love it. I use Demucs a lot and this program will certainly become my favorite very soon. Thanks a lot for implementing it and sharing it with the Demucs community.
I have a little suggestion: Please add an option that will make the program update the models to their latest version. Also, if possible, it would be great to personalize the output, such as having it in MP3 format at 192kbps, or in WAV format at 24bit, etc...
Keep up the great work!
Thank you for the support!
Regarding your first question, what do you mean update models to their latest version? Do you mean the Demucs models? You can actually add all of the currently released Demucs models to UVR by dropping them into the "Demucs_Models" folder. Just make sure to include the .yaml file and the application will automatically detect them!
Regarding the .mp3 and .wav options, I can definitely add those features to the next patch!
Thank you for the support!
Regarding your first question, what do you mean update models to their latest version? Do you mean the Demucs models? You can actually add all of the currently released Demucs models to UVR by dropping them into the "Demucs_Models" folder. Just make sure to include the .yaml file and the application will automatically detect them!
Regarding the .mp3 and .wav options, I can definitely add those features to the next patch!
Thanks a lot. I actually meant update Demucs whenever a new version of it is available.
I also noticed the you have added other models inside Demucs, including the 2-stem ones trained with UVR data. I have tried them all (2-stem and 4-stem) but couldn't notice any substantial difference between them. Which one do you think it's the best in terms of overall quality and fidelity to the original song?
They're very close in terms of quality. I think UVR_Demucs_Model_1 is a little better. One other thing to note is the Demucs v3 UVR models really excel when ensembled with the MDX-Net and VR Architecture models. Check out the Ensemble Customization Options for more information.
Regarding auto-updating when Demucs v4 comes out, that's all going to depend on how the Demucs teams codes it. I examined the code from Demucs v2 and Demucs v3 when I put all this together, as Demucs v2 was semi-implemented in UVR before this latest release and I can tell you it was a pretty big re-write codewise. So it's very unlikely auto-updates like that will be added. However, when Demucs v4 is released I will make sure it makes its way to UVR!
They're very close in terms of quality. I think UVR_Demucs_Model_1 is a little better. One other thing to note is the Demucs v3 UVR models really excel when ensembled with the MDX-Net and VR Architecture models. Check out the Ensemble Customization Options for more information.
Regarding auto-updating when Demucs v4 comes out, that's all going to depend on how the Demucs teams codes it. I examined the code from Demucs v2 and Demucs v3 when I put all this together, as Demucs v2 was semi-implemented in UVR before this latest release and I can tell you it was a pretty big re-write codewise. So it's very unlikely auto-updates like that will be added. However, when Demucs v4 is released I will make sure it makes its way to UVR!
I am trying with these settings for Multi-AI ensemble: https://imgur.com/a/Yoe3ZVI
Do you have a better suggestion in terms of settings?
Regards.
Thanks a lot @Anjok07 ! I added an entry in the README linking to this issue :)