itch.io

Thank you! The AI can generate arbitrary length music, but it does so in small steps so that you can have input on which generation to use. To make a 3 minute piece, you must click generate again after generating one section, then repeat the process multiple times.

Hmm.. it should be safe so I don't know why your antivirus is saying such. I am no security expert but, doing some preliminary research, the problem seems to be underlying in some quirks antiviruses handles certain things

I can hook you up with a refund, or as the header says, you can message me for $15 of credit on email.

Thank you, I am working on a tutorial right now so it should be out soon!

Ok, you are definitely above the requirements, and it looks like it actually does generate something before crashing, meaning its probably an issue with the handling after it generates. If it crashes before you switch tabs to look at the generation, it could maybe be an error with your drivers potentially. If it happens after you switch tabs, it may be due to drivers aswell potentially. Try using a different model to generate and if that doesn't work, idk. Also you may want to close out of any other windows incase you had them open before

Hmm... to be honest i got no clue, can you possibly send me your computer's specs?

Sorry... :(

This is due to the strict error handling in the midi loading section of the program, but for now you can boot https://signal.vercel.app/edit and just load the midi in the file dropdown, and just save it then reimport the midi into the program

Oh shoot, sorry about that. I just sent the friend request (i am hidude562 on discord)

This error means your GPU ran out of vram when trying to load the model files, so try closing out of any other apps (if you have any other open), and you may want to try restarting your computer too for good measure, then rebooting the software.

Apologies again, i seem to have mistakenly  forgotten to include fluidsynth in the build

New build up again, hopefully should work this time 🤞

Okay, the new build is up now, so hopefully upon redownloading it should all be fixed 🤞

Nevertheless, lemme try to get itch.io to refund you for this bad experience

Bruhhhh how did i mess up this badd

I am terribly sorry for all of this experience and lemme refund you rq (Itch.io must process it so it will likely be a few days)

I'm gonna try and debug this, i suspect i know the problem (it is an issue in the new update) but it may be upwards of a few hours.

I would say it really just depends on what the purpose is, so like maybe for a boss battle theme you might want to try something like "metal" or "dance-eletric" (note the spelling), or like for an overworld theme, you might want to try something like "film", maybe "soundtrack"

You can press generate again after generating a segment you like, and it will continue from that new segment and the previous. Right now it is indeed limitted to generating in small chunks

Openmusenet-2.11-L is basically an early version of the large size models, so it is not a small model

Openmusenet-2.11-M is the largest of the medium models, then Openmusenet-2.11-S is a smaller version, Openmusenet-2.11-3M is kinda a joke version, so i would probably recommend Openmusenet-2.11-M since its about 1/3 the size of the larger models and is still a somewhat reasonable quality (the first video in the samples is by it)

Yes, but probably not for a while because getting the app to compile to mac-compatible format has proven really difficult.

huh

Ah ok, I was able to replicate this bug with the gxscc app. I was able to fix it by importing it in this, then just simply downloading it again. You can also configure the instruments there if the gxscc app doesn't support changing of instruments.

Hope this works for you🤞

Hmm... are you trying to re-import the midi into the program by any chance? You may also need to also resort the channels in an app as with importing.

This is kinda buggy behavior so im gonna go ahead and add this to my bug list that im fixing and will fix this soon

The tracks are color coded. When you import your midi, is it all one color? If it is, i would recommend going into an app like signal.vercel.app, then try removing some extraneous tracks if there are any and then update the channels to be just in the first four (since the app separates  just the first four channels, then sets the rest to one channel).

Generating off of a certain length would be a good idea, when i'm able to create larger context models (For now though, it is probably not worth it because the current ai models really needs the context)

na sorry bro just download the demo tho if you just wanna try it out

Bruh thats so weird, i'll have to try and replicate that. Nice to hear that you finally got it to work though!

Ah, ok, you might be running out of ram then, mb for the inconvience bro. That means that you are probably limited to using the smaller sized models. 

Sorry bro, lemme get you with a refund rq (through itch.io so it might take a while)

Does your laptop have a GPU by chance? Generation should still work (albeit pretty slow) if you just have a CPU. The time/percent goes over 100% since it is just an estimate of the time remaining.

I think i've found the error (for some reason, i forgot to include the fluidsynth dll when building)! So try re-downloading it and it should (in theory) work. Seriously thank you for telling me this lmao i dont know how i didnt find this sooner

It actually does seem possible to do with pytorch according to this guys implementation but going off the benchmarks, it looks slower.

That is very strange and idk man. I'll refund you for that as that is obviously a big issue and im sorry for the inconvenience. It should still work on cpu. Can you detail me on the specs of your pc and were you able to generate in the demo version ?

(Note, the refund may take atleast some time because i have to contact itchio support)

You should be able to see/play multiple, you can move the red cursor with the right mouse button and cursor, and play/stop is in the bottom left

These messages are intentional, currently, there is no way to run the larger models with your gpu. The amount of notes the ai generates is automatically set to the max so it is not possible to generate more than the max

This error means you ran out of vram while generating/loading the model. So try closing out of other windows or try loading a smaller model like openmusenet-M or openmusenet-s

On vacation btw will be back like the 21st so i cant update rn if you have issues within this timeframe

This error occurs because you are trying to open a midi 0 file, which is (currently) unsupported. You need to convert the file first to midi 1 in order for it to work. I would recommend opening it in https://signal.vercel.app/edit then saving it via file > save and it should automatically convert it, so you should be able to open the new file then.

ohh ok. Generation probably did not freeze but if you don't have a gpu it will take like forever to generate (5 minutes probably for large model), which is longer than the duration of the demo. after 3 minutes the demo will close out. if you still want to still try the demo i would recommend going with a smaller model like the medium or small model.

That is a bit strange though for the fluidsynth error... Are you able to hear the notes when you play?

bruhhhhhhhh you gotta be kidding me

Can you send a screenshot or tell me what happens please?

Yup thats probably it sadly, windows 7 hasnt been supported in a while now so i think the newer versions of python dropped support for it. I think i recently upgraded to the latest python.

Its now been updated btw.

Yeah sure, sorry lol, I've been procrastinating updating the demo but I'l actually try to get around to updating this

*note* i recommend not using the new ai model for generating from scratch

Lol even on my school laptop it takes 15 seconds

Btw i might of gotten a bit carried away and trained an absolutely tiny model