Datasets:

4eJIoBek

/

PAIT-Downloads

like 19

im waiting for analogic model, but by stability and harmonai, they ll have increased sample length(90s) and 44100 sample quality, 15 seconds imo isnt serious and musicgen/audiogen will be somewhat better. also audioldm is big and will run on cpu very slowly. R u really so need it?

hi, thank u for ur answer!
hmm, there is an audioldm too, it is version 1 and has small models but having audioldm2 [not the large mode] will be good too.
yes i would like to have audioldm2 for its sound effects, it's be a nice thing to ur collection too
i think people will use this too, but please make this portable only IF U WANT IT TO and have free time to do it
thank u

ok, ill make it soon

thanks mate, cheers!
anyway, i found something very good if u r a stable diffusion fan, i'll make a new topic for it if it is ok

ok

for some strange reason it starts to sample something before the webui starts, idk why it is happening

hmm that is strange indeed... btw i just saw u uploaded fastsdcpu! thank u, it is a shame it can't be offline though, what could be the reason i wonder

transformers need to check some files in model repo even if transformers_offline variable is on

hmm i understand, thank u again! let us wait someome bring it in webui a1111

there is an extension for it!
sd-webui-lcm on github!
it supports fp16 too, can u check it out please? like converting the model to fp16

Seems like lot of folks are interested in lcm, i think someone ll make a normal integration of it into A1111 during this week, if it will be, then ill make another portable

in new update of fastsdcpu it can now be offline, if u need to update

Ow, ill update it today if everything will be fine

btw, could u pleaase check the audioldm [version 1]? it has a very smaller models i think they will be faster for cpu

https://github.com/haoheliu/AUDIOLDM

Ill check it when ill have more free time. If you dont want to wait, you can try to make a portable by yourself using this "toolkit" https://huggingface.co/datasets/4eJIoBek/PAIT-Downloads/resolve/main/%D0%BF%D1%80%D0%B8%D0%BC%D0%B5%D1%80%20%D1%81%D0%B0%D0%BC%D0%BE%D0%B7%D0%B0%D0%BF%D1%83%D1%81%D0%BA%D0%B0%D0%B5%D0%BC%D0%BE%D0%B3%D0%BE%20%D0%B3%D1%80%D0%B0%D0%B4%D0%B8%D0%BE%20%D0%BD%D0%B0%20%D0%B3%D0%BE%D0%BB%D0%BE%D0%B9%20%D0%BC%D0%B8%D0%BD%D0%B8%D0%BA%D0%BE%D0%BD%D0%B4%D0%B5.7z

oh thanks mate! but im really bad at this kind of stuff, please do it when u r completely free and have the time to do it

hey man! i managed to install audioldm, so please don't make a portable version if u dont want to, thanks

ow, how did you solved strange sampling for 200 steps before the webui starts?

i got audioldm version 1, it didn't have the audioldm2 issue

aw

Still soon™. Harmonai also doing some research about enhancing quality of autoencoder for now, so keep waiting

https://huggingface.co/datasets/4eJIoBek/PAIT-Downloads/resolve/main/%5Bexperimental%5DStable-audio-from-paper-t5-webui-portable-win.7z?download=true
this is that leaked stable-audio model, its experimental for now and webui generates only 48s for some reason, while model can do variable length (2-92s in paper), but it is still better than audioLDM2, so i think this discussion is closed for now

duration can be adjusted via varying sample size in stable-audio-tools\models.json file, ima too dumb to add it to gradio demo by self