search

rupeshs / fastsdcpu

Fast stable diffusion on CPU
MIT License
1.51k stars 123 forks source link

Web-frontend LCM-openVINO crashed hard #143

Closed euneuber closed 9 months ago

euneuber commented 9 months ago

Version 1.0.0 beta 24 Python 3.11.8 only web-frontend (qt5-gui does not work, see #140) mode: LCM-openVINO model: "rupeshs/sd-turbo-openvino" inference steps: 6 image width, height: 512, 512 number of images: 6 guidance scale: 1.2 on other options output format: PNG Text-to-Image: pos. prompt: "an AI robot ruling the world" neg. prompt: (empty)

After downloading the files for openVINO the python program crashes so hard, that even the terminal (konsole) dies! (I never had this with any other program before!)

If this is because of the settings I chose, please guide me!

rupeshs commented 9 months ago

@euneuber Please refer to the memory requirements https://github.com/rupeshs/fastsdcpu?tab=readme-ov-file#memory-requirements The guidance scale is 1.2 it will take around 15GB of system RAM for OpenVINO. Reduce the guidance scale to 1. For SDTurbo only 1 inference step is fine. Here is the latest release with Lora models support https://github.com/rupeshs/fastsdcpu/releases/tag/v1.0.0-beta.25