We shrank down Flux from 12B parameters to 8.8B parameters with minimal quality loss! Allowing for Flux to run on lower-end hardware and get a boost in speed.
You'll need to use this model with our ComfyUI node FluxMod.
https://github.com/lodestone-rock/ComfyUI_FluxMod
Check out the example images to see the difference in quality. In our view, it's effectively just a seed difference, but see how you go!
May your it/s be high and your VRAM usage be low!
Description
FAQ
Comments (3)
Is this meant to be paired with the universal_modulator.st model, or in place of, or some other thing .. I could use some explanation of where this fits in. Thanks!
I've had a great experience with this amazing tool. The speed has increased by about 30% (without closing any layers), and there's almost no loss in quality. Using the DEV version is fantastic! I'm really looking forward to future updates, and if they can apply this to the FT models, it would be the perfect technique!
Dont work for me, i can use the normale dev version its way faster then this "mod " here. it crashes the whole time with and BIG block of an text wall, idk what error cause i dont read it. i tried it a few times, there are around 3 minutes without doing any thing and then crashes and yes i youse the right files and the examplet workflow from github