>>70790155I think it's because my A1111 was old and had a bunch of junk in it that really slowed down my gen. VRAM usage was also significantly higher on A1111 for me. But overall, Forge has been working wonders for me for SDXL. Haven't tried comfyUI so I can't really comment on that.
>>70790416For me, loras tends to add more time to generations in general, even back in SD 1.5 days. A 1-2 second gen would take like 7 seconds with lora.
>>70790486I know that when I was on A1111 before moving to forge, I had TensorRT extension or something that cut gen time down to 1 second on a 1024x1024. But the load time it everytime I wanna gen stuff was pretty crap and made it not worth using.
