(older) UVR & x-minus.pro updates/news (2021-2023)
Moved here (just a few pages), reaching the Google Docs characters limit here.
_________________________________________________________________________
Online sites and Colabs for separation - the best quality freebies you can currently get
Refrain from using lossy audio files for separation (e.g. downloaded from YouTube) for the best results.
See here for ripping lossless music from Tidal, Qobuz, Deezer, Apple Music or Amazon.
If you don't have a computer, or decent CPU/GPU and separation is too slow on your machine using UVR 5 GUI, or it doesn’t work correctly for you, you can use these online sites to separate for free:
mvsep.com | Android | iOS | (lots of the best UVR 5 GUI models incl. various Roformers, and many exclusive models not available in UVR, and ensemble of these for paid users, now along with presets for advanced processing pipeline like custom ensembles and phase fixer)
The page by MDX23 code/Colab’s original author and creator of tons of models - ZFTurbo.
Rather refrain from using the site without registering an account, because you might be stuck in a gigantic queue for hours. Despite waiting the queue time, processing might take sometimes equally long like the timer from the queue.
If you register an account on MVSep, you can output in .flac and .wav 32-bit float.
Since 28.07.25, now a 32-bit float for WAV will be used only if gain level falls outside 1.0 range, otherwise 16 bit PCM will be used.
Also, FLAC now uses 16-bit instead of 24-bit.
If you have troubles with nulling due to the new changes in the free version, consider decreasing the volume of your mixtures by e.g. 5dB, and you won’t be affected, although it might slightly affect separation results.
Now a new overlap option was introduced. The default 50% means overlap 2, while 87.5% overlap is 8, and in the browser it is reserved only for Premium users. More about overlap.
If your credits are higher than 0, you have a shorter queue (also users using the mobile app have “a bit higher priority”), 100 MB max file size/10 minutes (up to 10 concurrent separations in premium and 1GB/100MB in premium). You can disable using credits for non-ensembles in settings, for the cost of a longer queue again. Shorter queues seem to be currently around mornings of GMT +2/CEST (9 a.m.) or even early afternoon or late at night, depends - sometimes the queue goes crazy long randomly, but if you don’t care, you can just set your jobs and download it the next day. Usually weekends have higher queues.
In February 2026 MVSEP introduced a limit of 50 separations per day for free users.
There were cases where certain free users were clogging the queue substantially.
“There won't be reset [time], it will just count all separations in the last 24 hours.”
Q: “Does the credit usage stack up the more I use a model? I used the becruily fullness model once and it cost one credit, I went to use it again on another file and it says 2 credits will be consumed. Is this intended?”
A: “Depends on how long the files are. Iirc 1 credit per minute.”
Selecting “Extract from vocals part” uses the best BS-Roformer models as preprocessor for the chosen model (currently BS-Roformer ver. 2024.08 - subject to change).
For Mel Band Kar dim_t 801 and overlap 2 is used, and for Mel Becruily inst/voc, Mel 2024.10, Mel Rofo Decrowd: 1101 and 2.
If downloading from the site is too slow try out e.g. Free Download Manager (Win) or ADM (Android) and/or VPN (although at least some are rejected now), or if you have premium you can use your credits to pack to zip the separation after your separation is done.
For some people it also helps to close all the MVSEP tabs while download has started.
As a last resort, use:
mirror.mvsep.com
de.mvsep.com
(just be aware that separations will be handled by machines available in any locations)
Batch processing with API and GUI - click (Mac). You can use MVSEP download links as remote URL in order to further separate the result (e.g. MVSEP Drums>drumsep for more stems).
Q: Is there a way to turn off the normalization when using FLAC?
It's annoying when you have to combine the outputs later
A: “No, if you turn off normalization, FLAC will cut all above 1.0
And if it was normalized, it means you had these values.”
FLAC doesn’t support 32-bit float, it’s 32 int, so normalization is still needed.”
So if your stems don’t invert correctly, just use WAV.
Q: How multichannel is handled by MVSEP:
A: librosa script which performs stereo downmixing (for 5.1 or 7.1 inputs)
Q: I convert a song using v1e+ and use phase fix, then do another conversion using for example Gabox V7 and use phase fix, if I go back to upload the same song using v1e+ it gives the stems instantly but if I use phase fix it will process again, in the past it would remember
A: This may be a temporary issue. Sometimes that server may be unavailable, then processing will start on another server.
Q: What’s “include results from independent models”?
A: “When you use an ensemble, you will also get results from each model of the ensemble and not only the ensemble final result.”
Q: What means “Disown Expired Separations” option
A: “we do not delete expired separation data (they are needed for analytics), but just remove your ownership from expired separations
We could have written delete expired separations, but wanted to be more clear about your data”
Q: “So I understand, all the uploads are kept, regardless of 'disowning' or not. So what is the distinction between disowning and not disowning? Is there one?”
A: No uploads are kept, just settings. If you disown, you won't see your expired separations
Q: I will need a refresher in terms. Separations are created from (audio) uploads. Separations are also not kept? Only the settings used, i.e. kuielab_a_drums, aufr33-jarredou_DrumSep_model_mdx23c_ep_141_sdr_10.8059, and whatever segment, aggression, vocal only, etc are selected at the point of hitting 'do it'.. in a manner of speaking..
A: Separation is when you choose settings and upload file, we just save the settings and delete file.
Q: How to use the same file over and over for different models in order to test them, but without reuploading the same file over and over
A: “You can use remote upload for this. Just use [the] link on file from previous separation. So you will not need to upload anything. https://mvsep.com/remote”
“we (...) updated server for China, Korea, Japan, etc. (North Asian regions), the previous server was falling apart due to disk load, now it should work faster for those areas, if not, let me know
We also closed translation requests, on its place we will launch revised Arena soon”
x-minus.pro / uvronlione.app / nextgen.uvronline.app (-||-, 15 minute daily limit for free, on the next gen 20 minutes, very fast, mp3 max 320kbps output for free (on x-minus by default; lossless for premium), some exclusive models for paid users).
The site is made by one of the UVR creators and models creator - Aufr33 with dedicated
Overlap 2 used for Roformers. At subscription level standard and above, the single song limit for Roformers was 20 minutes but probably it changed. For Mel Karaoke model, dim_t 256 and overlap 2 is being used
Ensembles work only for premium users in the nextgen when you enable:
=>Advanced settings>Home screen>Show post-processing algorithm selection menu
Then you can also pick Phase correction (phase fixer) in the home screen instead.
After you see the waveform when the separation is finished, then you click Ensemble button.
“Mel-RoFormer by Kim & unwa ft3 and some other models are hidden (at least in the non-next-gen). As before, you can find them here: https://uvronline.app/ai?hp&test” - Aufr33 (link for free users) but the nextgen site has some extra models too (it has three categories - recommended, test, legacy).
The model used for phase fixer/swapper/correction on the site was Mel-Roformer Becruily Vocal at some point (might be a subject to change).
FAQ (might be old)
Q: How come level 1 will be eliminated? Is it possible to leave it since I use this site very little and paying (2.79$) per month is too much and anyway 360 minutes of audio per week is a lot. I do 5/ 6 per week. It is a waste of minutes.
A: If you renew your subscription several months in advance, you can use Level 1 even after removal. In addition, once your subscription Level 1 expires, you can use it for another month for free (after removing it in February).
Q: What happened to Nextgen UVR Online? I used Discord exclusive lite 2024 plan, but doing an extraction I have a standard plan. Bruh.
A: That plan is long gone.
In the old version it still works. But the standard is better.
Q: But I need extract in flac to get lossless quality, because I'm really didn't want only in mp3
A: Then you cannot. “Well, I don't even know. I logged back in, and while I was doing the isolation, I was able to save in both FLAC and other audio formats. But when I tried to do the isolation again, it immediately switched to the default setting. Especially since how was I supposed to know it had already gone?”
___