This reminds me of the SDXL-era DMD2
#4
by Goghor - opened
Very good LoRA, thank you very much!
funny you should say that :) DMD2 was one of the approaches I considered but after research for my 4 step lora and time and quality reasons I chose Progressive Distillation and LADD-style GAN critic instead.
If I ever find time and GPU resources a 2 step version of the lora could use DMD2 but also preliminary research shows that Progressive Distillation and LADD-style GAN critic might hold up well (and do faster training) at 2 step than DMD2.