Download
1 variant available
Original web 【PCM】：https://github.com/G-U-N/Phased-Consistency-Model
Because SD3 Classification is closed, so I published this LORA here.
Notice, this is a SD3 LORA.
This is an accelerator tool similar as 【LCM/Turbo/Lighting/HyperL】.
SD3 needs 25 steps and CFG = 5 to generate one image.
With this LORA, you can use 8-12 steps and (CFG = 2-3 or CFG++ = 0.5-1) to generate one image.
OF Course, Same quality. AND saving time.
Comparing original, this version is 【small】and same stable.
Enjoy your AI work.