Thoughts? ;)

@lucidzachary473 Hey, I lost track of this model since I rarely do SDXL output lately. I went back to SD 1.5. Lemme take a look now, I sitll have it on my HDD.

So I tested it and posted some samples. It's quite good and the results produce nice, pleasing output with good anatomy. I think there is something 2.5D mixed in this model that gives it a fantasy look. As opposed to RealVIS that has a more real gritty look (see my other post from today). I have tried to use SDXL many times but I always come back to SD 1.5 for my uses.

I'll wager, like other 12+GB models, this model alone will eat all your vram, leaving nothing for lora or VAE and you'll be dumped into shared memory from the start... so gens, if they don't oom, will take an hour on anything less than a runpod running an H100 or something.... Figure, 13+ GB + lora + vae is pushing even a 30/4090's 24 GB...

Just tested on a 3060 12gb vram 36 seconds for 40 steps with euler a and i have model+loras+vae loaded so i would say for me it currently does

@Weatherby43 Maybe fix your settings? I can use 13gb model with LoRAs on my 3060 12GB. There isn't noticeable influence on speed. Sometimes I can use batch size 3 but often I default to 2