Daniel, the down point of Riffusion is that they do not offer comercial rights at present. I am more focused on Suno and Udio, just in case. I made several libraries with instrumental Riffusion. After sometime of using it (which is incredibly easy) the music seams generic. But though i am saying this, i am liking it a lot, and it is a good asset for AI video crafters. Thanks for your newsletter.
Hey Tita, that's good information, thanks! My focus wasn't on using these for commercial purposes, but that does sound like a dealbreaker for anyone wanting to use Riffusion that way. I think it's great we have so many options to choose from. I'm sure they each have relevant features for different projects, etc.
Udio has something the others don't, and I wonder if it has anything to do with being bad at prompt adherence. It feels like someone turned the temp way up there, I dunno.
But these are solid! I can imagine just asking an AI to write some instrumental background music for me to listen to while reading one day. That's kind of terrifying.
Fun songs. Run, Run, Hug! was done in Suno 3.5 instead of 4. You may have gotten a slightly different result from the upgraded model. There is a "remaster" option I think somewhere in the interface. I haven't messed with that.
I ran out of free V4 credits, so had to run with V3.5 for the last track. It's my understanding that this largely affects the fidelity of sound rather than song coherence/prompt-following, so it shouldn't have had much impact on the stuff I measured.
And yes, there's an option to remaster your pre-V4 tracks into V4 versions, but that again requires upgrading to a paid account...which I may consider at some point in the future.
Great piece! I’ve never looked at these tools before, perhaps avoiding intentionally, but glad to have you do the introductions. Agree with your take overall. Riffusion definitely the winner for me as well. Udio was the most uneven but I liked its punk sock song best by far. Just personal taste there. Again, thanks for the introduction to these tools!
The punk rock track was indeed pretty neat, but I had to subtract points for Udio ignoring most of the prompt.
Happy to introduce you to the world of AI music - would love to hear your thoughts if/when you end up trying them out! (They all have free tiers, by the way.)
Asked for a traditional country 12 bar blues in E about whiskey and women. My mistake was using the word “country.” Ha! Fun. I’ll keep playing with it.
I think overall that Suno has the best sound quality (in general) while Diffusion is pretty terrible in my opinion. That stated, I think, at its best, these sound like AM radio being played out of a 9v battery powered pocket radio run through a cheap earpiece into a Ronco Mr. Microphone (https://www.youtube.com/watch?v=hdJ7ng7Pchs).
But...
I do love that an entire song can be produced and there does seem to be a jump in sonics so it'd take much less tweaking in the mastering process to be usable on a wider scale than personal amusement or as a scratch workbook. For the time being, I'm thinking AI is better suited to creating single instrument or voice tracks that accompany human created stuff. It'd be easier to hide AI's inherent tonal flaws in a larger mix.
After all, I'm just an unsophisticated music consumer peasant who'd never come up with a description as precise and evocative as an "AM radio being played out of a 9v battery-powered pocket radio run through a cheap earpiece into a Ronco Mr. Microphone."
But yes, it's quite impressive (scary?) that we can now go from a simple prompt to a full song now. I do wonder how fast AI will progress in this arena and at what point it'll become palatable even to the most attuned ear.
My guess is June of 2026. I do wonder when or if these engines will have the option to download individual stems (instrument/voice tracks). That seems like the best way to attain higher fidelity to my high-falutin ears.
Daniel, the down point of Riffusion is that they do not offer comercial rights at present. I am more focused on Suno and Udio, just in case. I made several libraries with instrumental Riffusion. After sometime of using it (which is incredibly easy) the music seams generic. But though i am saying this, i am liking it a lot, and it is a good asset for AI video crafters. Thanks for your newsletter.
Hey Tita, that's good information, thanks! My focus wasn't on using these for commercial purposes, but that does sound like a dealbreaker for anyone wanting to use Riffusion that way. I think it's great we have so many options to choose from. I'm sure they each have relevant features for different projects, etc.
Udio has something the others don't, and I wonder if it has anything to do with being bad at prompt adherence. It feels like someone turned the temp way up there, I dunno.
But these are solid! I can imagine just asking an AI to write some instrumental background music for me to listen to while reading one day. That's kind of terrifying.
The first two results from Riffusion for "Some instrumental background music for me to listen to while reading."
https://www.riffusion.com/riff/4c074544-cab5-43c8-be31-2ad5084db638
https://www.riffusion.com/riff/bb7c18d0-7ef2-40d2-aaf6-5561d4194a46
You're welcome!
I've heard worse! also: better!
Were are we on full actor audiobook narration?
You mean something like this? https://elevenlabs.io/use-cases/audiobooks
Yes, now to figure out whether they can detect emotion, etc.
Speaking of, they literally just announced this:
https://www.linkedin.com/posts/elevenlabsio_from-today-spotify-is-now-accepting-audiobooks-activity-7298359757289979904-Lnnk
Coincidence? I think not!
The claim is that they're supposed to do exactly that, but I guess putting them to the test is the only way to be sure!
Fun songs. Run, Run, Hug! was done in Suno 3.5 instead of 4. You may have gotten a slightly different result from the upgraded model. There is a "remaster" option I think somewhere in the interface. I haven't messed with that.
Well-spotted!
I ran out of free V4 credits, so had to run with V3.5 for the last track. It's my understanding that this largely affects the fidelity of sound rather than song coherence/prompt-following, so it shouldn't have had much impact on the stuff I measured.
And yes, there's an option to remaster your pre-V4 tracks into V4 versions, but that again requires upgrading to a paid account...which I may consider at some point in the future.
Great piece! I’ve never looked at these tools before, perhaps avoiding intentionally, but glad to have you do the introductions. Agree with your take overall. Riffusion definitely the winner for me as well. Udio was the most uneven but I liked its punk sock song best by far. Just personal taste there. Again, thanks for the introduction to these tools!
The punk rock track was indeed pretty neat, but I had to subtract points for Udio ignoring most of the prompt.
Happy to introduce you to the world of AI music - would love to hear your thoughts if/when you end up trying them out! (They all have free tiers, by the way.)
https://cdn1.suno.ai/51c2c291-de99-4193-956b-4b449ca02e2c.mp3
Asked for a traditional country 12 bar blues in E about whiskey and women. My mistake was using the word “country.” Ha! Fun. I’ll keep playing with it.
Ha, well played! It's still surreal (scary? worrying?) that you can pump-out a workable song draft from just a short text prompt.
For sure!
I think overall that Suno has the best sound quality (in general) while Diffusion is pretty terrible in my opinion. That stated, I think, at its best, these sound like AM radio being played out of a 9v battery powered pocket radio run through a cheap earpiece into a Ronco Mr. Microphone (https://www.youtube.com/watch?v=hdJ7ng7Pchs).
But...
I do love that an entire song can be produced and there does seem to be a jump in sonics so it'd take much less tweaking in the mastering process to be usable on a wider scale than personal amusement or as a scratch workbook. For the time being, I'm thinking AI is better suited to creating single instrument or voice tracks that accompany human created stuff. It'd be easier to hide AI's inherent tonal flaws in a larger mix.
I'll defer to your professional judgment here!
After all, I'm just an unsophisticated music consumer peasant who'd never come up with a description as precise and evocative as an "AM radio being played out of a 9v battery-powered pocket radio run through a cheap earpiece into a Ronco Mr. Microphone."
But yes, it's quite impressive (scary?) that we can now go from a simple prompt to a full song now. I do wonder how fast AI will progress in this arena and at what point it'll become palatable even to the most attuned ear.
My guess is June of 2026. I do wonder when or if these engines will have the option to download individual stems (instrument/voice tracks). That seems like the best way to attain higher fidelity to my high-falutin ears.
Suno's had the "Stems" option since the middle of last year:
https://help.suno.com/en/articles/2425985
And it looks like Udio does as well:
https://x.com/udiomusic/status/1816506741158469824
That’s pretty cool and I love that it’s the one with the least degraded sound. I’ll need to play with it and give you a comprehensive review FWIW.
Nice!
I'd love to hear more from someone who understands this stuff deeply rather than my "pretty song go boom boom" surface-level takes.