Closed KintCark closed 3 days ago
In LCM-LoRA mode you must provide the LCM-Lora model ( It will generate image CFG <=1 ). If you are using a negative prompt then a higher CFG needed.(Its already documented in the readme)
Oh but when putting the cfg to 5 with a negative prompt it creates a burnt image. Maybe u could add an non lora mode where we can use single file safetensors without lora lcm like a non lcm. Tab for original models.if u understand what I'm trying to say:)
Original models needs 20 to 50 steps to generate an image (slow).Original models use CFG 7.5 or higher but it will make image generation process slow.
Please add an none option for the lora selection higher cfg needs this using higher cfg with an lora makes burnt messy images