@stephenajason Have you considered/found any use cases for an H3 PUSA version?
Had fun w/ you Wan versions but I'm very much a layman w/ the coding/training/potential application to the newest crop of models (LTX/H3).
@WastewaterSCAN Wonder if graph starting to drop at the end there correlates w/ ~2wks post Pfizer/Moderna shipping this years updates? Seems close/plausible.....
@ArtificialAnlys You must be making bank off the $$$ you get for pushing this shit....open weight+fine tune and they go closed/you pimp for them?
Lame.
@andomort@ArtificialAnlys Well they WERE open weights (H3) which they fine tuned and are now trying to profit off of...
And these guys pimp all the way.....yippe.]
I guess finetuning an open model, renaming it, and making it closed source is the new thing now. I wonder how much you have to pay AA to get your closed source fine tune on there.
@EtherCoins@aisearchio Exactly this - tested it on a 6000 and yea if you going 1080p for 20sec it'll cut time down, but if not if your doing typical videos for kicks.......It's not the next coming of causvid or lightx2v........had Fable make a node, tried it, and back to cache......
@FurkanGozukara@xieenze_jr Kinda like your ability to say thank you is insufficient considering how much you earn off others free work there, eh captain?
@kohya_tech I'm surprised LLMs don't serendipitously lead us to more of these things... IE: Don't they train making associations between billions of things? You'd figure they're just spurt out stuff they assume we know already cause it's something at this point in time we should just know..