merge flux-krea, image created with the addition of lora FLUX.1-Turbo-Alpha which seems to shorten generation times 10 to 15 steps, this is a first test in the testing phase, I use euler simple and dpm++ sgm uniform seems to work.
https://huggingface.co/ByteDance/Hyper-SD/blob/main/Hyper-FLUX.1-dev-8steps-lora.safetensors
image generated with sd forge
for nf4
If you are short of memory, I recommend that you enable the Never OOM Integrated option on SD Forge.
Description
for small configurations advise for sd forge in 10 steps with https://huggingface.co/ByteDance/Hyper-SD/blob/main/Hyper-FLUX.1-dev-8steps-lora.safetensors
FAQ
Comments (6)
Is this actually Krea based? or is it .1 D?
merging with krea and my nunchaku model as well as the loras hyperflux turbo and badmilk
vince11 Do I need the LoRAs or they are "merged" into the model?
So you can merge the weights of Krea, with the original flux?
wooba001466 Hello, yes I tinkered a little with Forge and for me it seems to have worked, I don't know, maybe a stroke of luck, I try a lot of things, often it crashes and sometimes it works, have a good day
entrypointjip the loras are merged into the model, I add lora:Hyper-FLUX.1-dev-8steps to a value of 0.20 for krea and 0.12 for flux dev, this allows the steps to be reduced to 8 to 10 steps