FIX FP16 Errors SDXL - Lower Memory use! --- sdxl-vae-fp16-fix by madebyollin - v1.0 Review
Loading Images
Typical RAM/VRAM usage during generation.
Win 11 / A1111 1.6.0
SDXL model + this VAE, 832x1216, batch count: 2, batch size: 1, 30 steps + highres fix at x1.5
Launch args: --medvram --xformers --opt-split-attention --gradio-img2img-tool color-sketch --api --update-check
OMG i'm not depleting my whole RAM anymore !
No more absurd waiting time between basic generation steps and highres... (30 sec x batch size so... about a minute for a batch of 2...)
No more ultra laggy whole system because of 98% ram use...
Possibly no more websocket connection timeouts...
Faster generation on top of that.
What can i ask for more ? Is this christmas already ?
Thank you so much for that, and the launch argument tips !