I Compared Suno And Udio Ai Stems, So You Don't Have To!

[@VampyressVA]

The whole AI should be re-made to think in terms of: generate one instrument, then another, then another, then vocals. Master the whole thing. Save current mastered tracks separately. Present the result as the *combination* of all of them. But currently it's like this: generate instruments and vocals, merge them together destructively, master the whole thing, save destructively. Oooops, now we have all sorts of echoes and harmonics floating around, so to separate them at this point will leave lots of bleeding harmonics and frequencies here and there, specially when there are panned vocals and stereo effects. It's the same as trying to separate individual ingredients of a stew after the stew was all cooked together. You're going to get bits and pieces of everything on everything. Stemming right now is no better than center-cancelling or the "karaoke" function.

[@ChrisUnlocksAI]

You are spot on! I definitely agree

[@danthony5455]

Well Said!

[@sub40hz]

This is what I told them two months ago. If AI can create a track, then AI generate actual stems, rather creating a track and using stem separation tools. The latter is lazy, and undesirable due to artefacts and quality. The AI could also provide its mixing chains and midi.

[@VampyressVA]

@@sub40hz That would be ideal! What they're offering currently is the solution that wouldn't require them to train a new model with very different instructions. I guess for now they're just not willing to risk it. But I'm sure what you described is in the plans... (one can hope)

[@sub40hz]

@@VampyressVA the way I see this AI stuff is that it’s there as a collaboration. Comes in with an idea, you then take the components, replace or add you’re own and then you do your human bit with structure and arrangement. That way it’s a true joint effort.

[@justincanton]

This feature will never be want we want, until they have the power to run models that can reverse engineer the initial generation into properly separate audio files. Rather than the stem separation being an after thought of a music file. Thanks for testing these out, It is exactly the quality I would expect for now without them doing what I mentioned above. 😊✌🏼

[@MrEnagee]

Exactly, , Thinking of getting udio but i dont think the stems will be good enough to use. RipX will probably do a better job of the 2trk. Thx for the vid

[@ChrisUnlocksAI]

Rip x has the advantage of turning the stems to midi. That's an amazing feature. Both these comments are spot on. I hope it gets better. I was not impressed. However I was able to replace the vocals drop them in over the crappy instrumental stems and in the end it sounded great. IDK😂 Thanks for watching.

[@GodAtum]

@@ChrisUnlocksAI I tried that but the MIDI is awful. Even with a simple piano the MIDI it produces misses notes, gets wrong notes etc.

[@hitmusicworldwide]

​@@GodAtumAre you talking RipX?

[@MelodiiCuAI]

I like the new premium options but I still work my way: take a song I like, extend from the time I want, vocal remove, work everything in my DAW an repeat if it's necessary. I managed to have nice songs in that way.

[@GiovanniIglesias]

Thank you, brother! This is really helpful for those of us music producers exploring AI for the first time.

[@ChrisUnlocksAI]

Awesome, glad it proved helpful. As a musician / producer myself I have really enjoyed working with Ai. You should check out the audio input feature for both Suno and Udio. Pretty amazing stuff. I did do a video on that as well if you're interested. Thanks for watching.

[@marcosazambuja]

Amazing enterprises, Suno and Udio. Too bad that they rely their success on a massive amount of stoled music. But after they face UMG/Sony/Warner lawsuit, other companies will take the same technological path trained only on allowed music, and even music that is created specially for this purpose. Google and KZbin are trying this way, and hasn't released their models openly to the public because their AI generated tracks aren't completely cleared. They have a lot to lose, and on the other hand Udio and Suno just see gains on their strategy: they intend to grow their catalog as fast as possible using consumer uploaded music/instruments, in order to close a deal with TikTok and Instagram. That's their vision: sell music to be used by this social media giants, without the royalties payed to the major record labels/companies.

[@DaveCoolMusic]

Keep doing what you love. You are doing a great job and your videos are amazing!!!!!

[@ChrisUnlocksAI]

Thank you so much!! I really appreciate the compliment.

[@sorenandrews1078]

I'm a music producer and I constantly use high end vocal and song editing software. (RX8/9). There will NEVER be a Stem Separator that gives us 100% clean, no artifact, recordings. The reason for this is pretty simple, Reverb/Delay. Even if you try to remove Reverb/Delay, you're going to get artifacts

[@ivanriobla]

that rx is left behind compared to many of this new AI steam sep. i think rx was the best, but ai is changing by the dy so i do believe we have better options

[@musiclife_indonesia]

Thats right

[@BeyondPC]

I tried splitting a couple songs but I could still hear the vocals in the instrumental part. The vocals seem clean enough but the music is contaminated with whispers.

[@ChrisUnlocksAI]

Definitely agree. It's a bit frustrating. Thanks for sharing your experience.

[@rula42]

7:36 I had this "yeah" moment too. Udio Rise and Shine II sounds great

[@ChrisUnlocksAI]

It was a Banger! I may finish that one!

[@SoniaBenstead]

Hi, good video. Do any of these platforms generate sheet music for the songs you create? So you can give to performers to do your songs. Ie chords for guitar and melodies for piano?

[@Carmen-mp3je]

Great question… I would love to know if it is possible for both me playing live or my musicians or my students

[@Ironbarkbob]

Thanks for running through the process. Great video.

[@ChrisUnlocksAI]

So glad you liked it. Thanks for watching.

[@Machinesongs-v2d]

So, Udio still has better sound quality while Suno generates better melodies. Neither have stem separation that is an improvement on what can be done with other sites/apps. Disappointing but thanks for taking the time to compare them..

[@ChrisUnlocksAI]

You're welcome. And yes I agree. It was pretty disappointing. However in the end you still can just replace the vocals and it sounds pretty good. But if you are looking for just the instrumental it's a wash. Thanks for watching.

[@thomlev5217]

can we stem every instruments in suno or just voice instrumental?

[@Fuchsia_tude]

What's with the 5:14 of black dead air at the end of this video?

[@ChrisUnlocksAI]

I'm not sure. I will look into that. Thanks for letting me know.

[@EclesysGalaxy]

So the stem is basically the same as running the original audio through mvsep? You can get pretty close results for free through mvsep if you know what to use over there.

[@dennisgordon7767]

thank you.

[@wemoood]

Hi Chris, hope you are doing well brother. Quick question bro, which one has the best sound quality from this 2 overall? udio or suno? Would really appreciate your response!

[@ChrisUnlocksAI]

Hello, It really depends on the style of music you are wanting to make. If it's an older style like Motown, classic Americana, old country Udio is 100% the way to go. If you are wanting anything modern. Suno alll the way. Hope this helps, and thanks for watching.

[@wemoood]

@ChrisUnlocksAI Appreciate your response dude. I was referring more at the bitrate quality, I felt like suno doesn't do se well compared with udio

[@Im_Camping_at_B]

Could you use these for gaming? Warzone?

[@ChrisUnlocksAI]

If you mean to create music for gaming. Yes absolutely. Hope that helps.

[@cfig81]

Thanks for the vid, bro. Do you know why Udio change from 32 bit audio to 16 bit? instead of upgrading they downgrade.

[@ChrisUnlocksAI]

I don't know why. They do some really great things. And sometimes not so great. Thanks for watching.

[@packedtrivia]

what program did you use to change the vocals

[@ChrisUnlocksAI]

Audimee Ai vocals. There is a link in the video description. Hope that helps and thanks for watching!

[@GVEMiKE]

All of these tools have AMAZING potential but to be honest what they can do is MORE than enough to export an idea out and recreate with real musicians and vocalists to be honest. You could event take a lot of the drum information and turn them into Midi files so this is only the tip of the iceberg tbh... Great video bro!

[@alfonsoagren414]

Have you tried to do the stem separation directly in Logic Pro and compared? I think there is some problem with AI music because it is creating the sounds differently than a recording. When i separated dtems some guitarrs as an example sounded very eeird and cut iff at points. Probably because the AI just generates the guitarr where it is hearable in the mix. In a recording the guitarr would ring out notes even if it woukd be more or less unhearable in the mix. AI turns it off. We record multi track while the AI treats the whole thing as a single instrument in a way is my guess. Cheers!

[@huevonesunltd]

Here's the thing with Music AI , it's getting massively ignored by both hobbyists and researchers in comparison to the other fields, it's like the niche of AI, most people out there are extremely busy with improving Text generators (when it comes to LLMs models get obsoleted in a matter of weeks, its going crazy fast) and Image generators, , and then Video, and then voice and then music, in that order. Right now there's not many alternative for running local music AI and that's part of it. Once people really get into it, you will see there will be a sudden "revolutionary" jump in this music AI stuff, it happens often with text and images because both the community and big companies are really invested in those, for music its being very slow to the point its being "ignored" pretty much but one day its gonna get crazy good overnight once there's enough interest in it.

[@MarquisVonMonster]

Well if that's the case people interested in AI music production need to jump on NOW. Before the change in audience interests gets bloated with too many bandwagon jumpers and the music industry starts throwing all their money into AI artists. Best to learn it now and be ahead of pack rather than try to play catch up after the race really starts.

[@dn-anonymous]

What is your screen recording software? Handy tool.

[@ChrisUnlocksAI]

I use Filmora wondershare video editing for my screen recording. Super easy to use. Thanks for watching.

[@dn-anonymous]

@@ChrisUnlocksAI is the free version adequate for this, or do you have to pay? It's just that there are so many tools to get started with video editing, and I can't pay every subscription.

[@musicingflowing]

I got much better results using UVR5 (which is also free, and comes with Matchering mastering capabilities built into the desktop app), than with Suno stems. But in the song I've been working with, I've still been getting the same 'artifacts', regardless of which model/combination of models I've used thus far with UVR5.

[@HosannaSookra]

Just download your songs and put it into audacity to remove vocals and keep instrumentals. Works like a charm.

[@ChrisUnlocksAI]

I'm a Mac guy never used audacity. I will try it out. Actually thinking about doing a stem comparison with several brands.

[@philjohnson725]

Audacity's vocal remover is dismal at best. Tried it and it's just as crap.

[@hamsikjh719]

Can we remove the singer's voice and sing instead?

[@ChrisUnlocksAI]

Yes if you download the stems and put it into a DAW you can record your own vocals. Hope this helps. Thanks for watching.

[@dirtydevotee]

9:00 => We finally get to the actual "stem" portion of the video.

[@ChrisUnlocksAI]

True. I did have to generate new songs and try both programs, that took 8 minutes and 59 seconds

[@VampyressVA]

There's also the "Clarity" parameter in Udio. Has anyone tested it in different strengths to verify if stemming becomes more precise?

[@ChrisUnlocksAI]

I'm going to. Thanks for the suggestion! And for watching!

[@VampyressVA]

@@ChrisUnlocksAI Don't mention it - thanks for the video, great analysis!

[@ghost-user559]

Too much and it sounds “too clean” too little no it sounds too muddy. I like to start low and step it up for various parts of the song I want to emphasize more

[@VampyressVA]

@@ghost-user559 Interesting, thank you! Have you tried to stem the high clarity and low clarity parts and compare the quality of the stems?

[@ghost-user559]

@@VampyressVA I’m using the daily free credits so my experiments are slow and I learn as I go. If I had a plan I would definitely do more detailed testing, but from what I have sound the lower fidelity doesn’t allow for clean stem separation, but it sounds more “authentic” because in a real recording the sounds all blend together. But with the high clarity you get great stems, however the track sounds more artificial. The sounds are too perfectly separated. For more vintage sounds I would do a middle ground clarity, but for something like a modern orchestral soundtrack or a clean vocal mix, or smooth jazz then clarity to the max gives great stems.

[@a.l.1044]

What you're here for starts at 11:15. Don't thank me.

[@ChrisUnlocksAI]

I'm going to be time stamping my videos so you won't have to do that.

[@Sneakycat1971]

The stem separation is not worth it but can be convenient with certain songs. I want a feature where you can keep the singing style and generate the music or vice versa.

[@ChrisUnlocksAI]

Yes! I agree. I feel like they were both rushing to get stems out. Hopefully they can focus on quality instead of innovation. Thanks for watching.

[@chickendog9957]

I have done a couple of months of Udio. I wanted to subscribe to Suno but I can't seem to get an answer to a question. Maybe someone here could answer it. When I tried the free credits for Suno my download was a low bitrate MP3 file at 192kbps. When subscribed to Suno does is give you the option of 320k MP3 or WAV??

[@ChrisUnlocksAI]

Yes you get Wav files as well. And now the 2 stems options as well. Hope that helps. Thanks for watching.

[@chickendog9957]

@@ChrisUnlocksAI Yes Thanks!

[@VampyressVA]

You get WAV, but lower bitrate than Udio. IIRC, Suno gives you 48 kHz 16 bit, as opposed to 96 kHz 16 bit for Udio. Or that was true up until a few weeks ago - not sure if recent updates changed that.

[@chillwalker]

As both tools are not "composing linear multitracks" the "Stems" besides the vocals cant exist.... Try to seperate Suno or udio instrumentals and you will see its not based on recordimg instruments, its more than a "Film Set" of a Song. a momentum with "Fake" instruments...know what I mean? So if you are inspired by both AIs..do your a favor to be better then all the rest: Produce the inspiration agai by yourself. It also prevents your music and work not sounding like AI. When you experiment with all thw tools like I do agressivly for the last 8 weeks..you will know what I men. There is a point when you hear the training cause there are elements always repeating over all styles. Thanks for the vid. As I said before: I am too lazy to make all of my research to content... So keep on doing it for both of us, Chris!

[@ChrisUnlocksAI]

Ha great comment. You are 100% correct. I will do my best to keep making videos so you can stay on the chill side of things haha. Thanks for watching and the comments.

[@chillwalker]

@@ChrisUnlocksAI my pleasure!

[@oraldefication]

Dude... the udio songs sounded like crap. Vocals were all over the place. But I guess it comes down to personal preference.

[@perrottadivino]

so udio-130 has no difference from udio-1.5 audio quality or control? its just for better lirics?

[@Knk082]

If you want a cleaner sounding stem, just re-upload the file in Udio and remix it as close as possible to how it originally sounded. I experimented with an acoustic guitar sound I liked and it came out sounding better.

[@Mopantsu]

Udio should be doing that during the stem seperation process. That is an interesting workaround though.

[@Dragon_Slayer_Ornstein]

The before Udio voice is better than the tweaked one.

[@galefraney]

Do you use Logic Pro, and if so, have you compared these stem separations to Logic Pro's new stem separation feature?

[@galefraney]

oops, spoke too soon ... I see you do use Logic Pro later in the video 😄

[@ChrisUnlocksAI]

I have not tried the stem separation feature in Logic yet, I am a fan of some of there other new tools. I may do a stem separation computation between all of them. Thanks for watching.

[@VampyressVA]

Good question, I'd like to see a comparison, too. Everyone raving about the AI stemming in Logic Pro.

[@smdemmery]

I’ve tried both Logic and Udio stem separation and to my ears there’s very little difference. Early technology it’s getting there but several years away from the product we all want I’m sure 😎👍

[@VampyressVA]

@@smdemmery Interesting! Thanks! I had hopes that Logic was better at stemming.

[@TeeCee-qq4ev]

I used Udio to make a bunch of songs. The first 25 or so was incredible. They didn't miss an "I' or a "T'. Them joints were smoking with them backing tracks and my lyrics Then it just went crazy. Wasn't reading nothing I wrote, just spitting out what it wanted. Man, I used up about four thousand credits on ish that had nothing to do with me. I'm about my lyrics, not Udio, jamming.

[@jumpstar9000]

w00h00! Well. Like you said, the results are pretty disappointing, but as you also said, this is the worst it is going to be. It's nice that they are working on it. The Udio vocals sounded pretty decent after Audimee. Anyway, things are moving forward. Thanks a lot for the heads-up!!

[@ChrisUnlocksAI]

Thanks for the support. Hopefully this time next year it will be amazing

[@chillwalker]

Funny thing: UDIO is very strong in everything with Reggea and dancehall vibes.

[@ChrisUnlocksAI]

Yeah I'm loving it!!

[@lNowHerel]

Just get ripx and it will give you the stem ripper that you want! And it breaks down stem instruments further than all other programs.

[@PigeonKicker1234]

These AI's love using the words melody and echo

[@ChrisUnlocksAI]

They do!!

[@giannibene2465]

Bona parte del video finale e tutto nero. Un saluto, Gio'

[@ChrisUnlocksAI]

Thank you, I will look into that. Thanks for watching.

[@CornelioAustin]

Wait those Suno vocals are actually great. Udio was weird

[@leksbeet662]

I wonder what they do and why... They create several lines of tools. then combine them into one mix to create an mp3 file. Why not provide original tracks of individual instruments? Taking resources to separate stems from an already mixed track is stupid, in my opinion. But yes, saving unmixed tracks requires additional space on the server, maybe that's the problem...

[@ChrisUnlocksAI]

Definitely could be the issue. I hope it gets there eventually. Thanks for watching.

[@RandomVelocity]

Interesting to hear udio better than Suno. I just subscribed to Suno and I think it’s pretty phenomenal but in the mobile app I can’t find anywhere to get stems I would just use another separation tool to do that. Maybe it’s only available in the actual full web version, but now I have to try. Udio if it’s better quality than Suno

[@ChrisUnlocksAI]

Udio has better vocals than Suno overall. However Suno seems to do better with song structure. Both are really good if you keep working the prompts and rolling new creations. I subscribe to both. Hopefully you can find one that you love. Thanks for watching

[@RandomVelocity]

@@ChrisUnlocksAI I’ll have to try that too then. I’m experimenting with uploading my own music to get some more ideas as to where to go with it. It’s pretty good sometimes. Not sure about using any melodies though. Don’t really know if this system gets it if the melodies come from existing songs which would violate copyrights. I hear the record industry is planning a law suit basically because they want to do this so they can make all the money. Typical greedy and corrupt music companies.

[@Mopantsu]

The AI should be looking at the stem and 'recreating' the audio just like img2img in AI art. That way even though it won't be exactly the same audio from the original it will be a 'recreation' instead, but with a much cleaner resulting audio file. It should be able to do this to the point that the differences between the original stem and the 'recreation' are indistinguishable in all but the quality of the newly created stem. If AI art can reproduce a new image from an original image then something similar should be possible with audio. It's only math afterall.

[@Fatman305]

Spot on. When I hit Remix in Udio, with Similarity

[@Morenob1]

I must say Suno's ai vocals are very good made and really cover your track perfectly but the sound quality isn't so good, it hides too much in the track. I made a remix of Arianna Grande's song and let Suno finish the track with it's own lyrics. I've let ai seperate the vocals from the track so I can use that vocal again and replace it with a higher quality of Arianna Grande's vocals, but the words just aren't clear enough sometime. I see downloading stems is just ai separating the tracks.

[@hoaha6343]

we should share AI music so people can recognize it. I saw many people get praised by making song with simple prompt but no one know that's AI. It will also make true artist get more recognization and job because simple and repetitive industry works is now so cheap and easy with AI, so we don't have to compete with some dude who pirate DAW and sell shit beats for 20$

[@ChrisUnlocksAI]

I think it will be a requirement eventually. KZbin already requires it. Thanks for watching

[@theandroids]

Don't know why they call it stems. I nearly signed up thinking you would get actual stems. Not an acapella and instrumental. Honestly suprised that with AI doing the stem seperation thing for a while now, this isn't standard. I mean, should be easier since the AI is the one actually generating the music.

[@gu9838]

so its not FULL On track isolation? that sucks. like you said concsidering ai is making it you would think they would be able to have CLEAN source tracks. lol but neat i guess

[@ChrisUnlocksAI]

Yeah. Hopefully one day they can get us clean tracks. Thanks for watching.

[@thesvalker3720]

Обе сети сгенерировали практически один тембр голоса и стиля его исполнения

[@BabylonBaller]

Sad but their both using the same algos the DJ apps are using like Virtual DJ and Serato. I was expecting clean stems as well, like studio quality. These are a post process, nothing special

[@ChrisUnlocksAI]

Yeah. I agree. In the end I was able to replace the vocals, and it sounded good. However what if you wanted the bass from song. That's not happening. Hopefully it will get better. Thanks for watching.

[@JOHN.Z999]

I love SunoAI, but I must admit it is falling behind in terms of sound quality and music editing tools. They urgently need to release a better model. The best melodies I've created are from Suno, but the vocal and sound quality are severely outdated. I hope they release something new to compete with Udio, because otherwise, they will quickly lose market share. 😢

[@ChrisUnlocksAI]

I am right there with you. Hopefully they make some more progress with quality before adding new features. Thanks for watching

[@sgtjoe2008]

for language in my country, suno does a lot way better than udio. i'm curious about other languages if who's more accurate among these two.

[@thcredeye4016]

Udio is so much better than suno that i kinda do not understand why people use suno... unless you wanna make a whole song in 30 sek for fun, without controll

[@BoldCarnivore]

I just tried the Suno one and it is garbage.

[@ChrisUnlocksAI]

It's definitely got some issues

[@pheargoth]

Because both of those models use diffusion to create the entire track at once, and don't build the track from component parts in the same way a human does, it's understandable that they use similar tech to that which UVR5 uses to separate stems.

[@Twizzforsberg]

Didnt like udio at all suno all the way. But I need to learn how I can get my voice to a track

[@ChrisUnlocksAI]

Are you wanting instrumental music so you can sing over it?

[@Twizzforsberg]

@@ChrisUnlocksAI I dont know how it works. Can I write lyrics and then they use my voice and fix it or do I need to sing and ai fix sonit sounds good?

[@nickmack7282]

getting nice stems from crappy generations or crappy stems from good generations, ... why suno isn't just giving out what bark and chirp are doing instead of using a cheap steam splitter

[@ChrisUnlocksAI]

You have great points. I'm not familiar with Bark and chirp? Can you share more? Thanks for watching!

[@CozmoBeregofsky]

Great video - below average music, not to mention the borderline plagiarism.

[@phantasiaentertainment2170]

both stem solutions are bad. they need to retrain and advance their models so that it builds the content it creates as layers.

[@ChrisUnlocksAI]

I think they will get better. Thanks for watching.

[@marcduchamp5512]

Dunno why a lot of generated songs sounded like it was sang by a gen Z or X without much life’s experiences LOL

[@epytaffskitchenstink]

I tried to get Udio to extend songs and it just wouldn't do it properly not even once. So I switched to Suno and you can dump a whole song in there

[@Michael_Lak]

Wasted time spent on these features. Stem separation anywhere still not ready for Primetime. "Other" separation equals fail. Each song should be made up of composed components so stem separation not necessary, its already separated.

[@ChrisUnlocksAI]

I hope that's something they can work on in the future. Thanks for the comment and watching.

[@philjohnson725]

Stem separation on Suno and Udio is crap. Period.

[@joepipino]

So who will play this on the radio etc so you can make a living as a songwriter? Or do you become a glorified music editor forever? There goes another job taken over by AI.

[@ChrisUnlocksAI]

I don't think so. This will just be another tool to use. Thanks for watching.

[@roxylosbots]

I think Udio is much better, I've realized these albums with free version and the sound is good & music beautiful: kzbin.info/aero/PLcdOknWRyqNZvXSFf0vuzqsAD6hJvaln9&si=44Hm2_qJWWrH-SSZ

[@marcduchamp5512]

Gawd awful vocal sound quality stems

[@MochaMediaStoryteller]

CAN YUO START PRODUCING SHORTS SO I’M NOT INTIMIDATED BY SHOW LENGTH OF 30 MINUTES FOR THIS INFO...ijs

[@ChrisUnlocksAI]

That's very fair. Yes I will work on that. Thanks for the input.