Bonus: Pixart 900M "Mixture of Experts" Edition!
Added 2024-07-21 14:33:02 +0000 UTC
Even more PixArt! Things seem to be dropping every day, and now we've got a 2-stage pixart. While you can use each model individually, it's better to go with the "refiner" style workflow and let the first stage do it's job with composition and colours, with the second stage adding more clarity and detail.
Models can be downloaded from:
https://huggingface.co/ptx0/pixart-900m-1024-ft-v0.7-stage1
https://huggingface.co/ptx0/pixart-900m-1024-ft-v0.7-stage2
Example workflow attached below - still trying to find the best steps, samplers & schedulers but this config seem OK to start with. Have a little experiment and let me know if you find anything fun :)
Comments
it's hard to explain, but ffmpeg hits my cpu, nopt my gpu, when it's separating all the frames from a movie when it first starts processing. I was just wondering if there's any truth to optimizing nvidia / windows settings these days. I know back in the SD1111 days, we were duct tape fixing things :D no worries not raelly a question and i cant believe u do tech support on here....<3 I wont put that on ya
NetDude
2024-09-10 22:33:35 +0000 UTCSorry, I'm not getting the ffmpeg thing?
Nerdy Rodent
2024-09-10 12:06:44 +0000 UTCmy computer can/will hit 85 celcius if ffmpeg gets going on the CPU. For that reason, I always keep threads at like 6. Is that even correct? I assume threads are my CPU in most situations (Face Fusion or Topaz Labs). I even tell Blender to avoid my CPU but I always go away to get a spot o' tea, love... I come back and my computer is a friggin hovercraft trying to cool itself down with a 420 arctic/3090 :+ I know that some apps refer to the GPU in threads, and I get that. but is there value in lowering the threads + telling Nvidia not to do the whole unload to my 5950?
NetDude
2024-09-08 00:32:19 +0000 UTCThanks - I actually managed to find it and was coming back here to say ignore this!
Stephen Hilsdon
2024-07-31 18:56:34 +0000 UTCT5 is the same as the regular PixArt and the files are also in their repo - https://huggingface.co/ptx0/pixart-900m-1024-ft-v0.7-stage1/tree/main/text_encoder - place them in your "ComfyUI/models/t5" directory. If you want to use different paths, there is more info for those custom nodes at https://github.com/city96/ComfyUI_ExtraModels?tab=readme-ov-file#t5
Nerdy Rodent
2024-07-31 16:36:52 +0000 UTCSorry silly question - where do I get the T5 model from and where do I save it?
Stephen Hilsdon
2024-07-31 10:38:29 +0000 UTCIs that set on auto or cpu? I run it on cpu myself
Nerdy Rodent
2024-07-22 17:47:37 +0000 UTCi'm getting the following error - Error occurred when executing T5TextEncode: Tensor on device cpu is not on the expected device meta!
Brandon Waters
2024-07-22 13:25:19 +0000 UTCThanks!
Richard Gates
2024-07-21 23:14:53 +0000 UTCFiles -> Transformer
Samuel
2024-07-21 22:45:12 +0000 UTCAm I missing something? I don't see the models on Huggingface. I only see a file called optimizer.bin and it doesn't work.
Richard Gates
2024-07-21 21:47:34 +0000 UTC
