The generation speed of AI paintings will be restricted by the following factors: torch version, transformers version, CUDA version and cuDNN version.
Users of non-40-series graphics cards should use the latest modpack for best speed. The v3 integration package has been updated to torch 1.13.1, CUDA 11.7 and transformers 0.016, so no other changes are required.
A method to revive the Stable Diffusion WebUI at full speed. The maximum image generation speed is 30+it/s.
Users of 40-series graphics cards need to replace cuDNN to get full speed.
First download cudnn.
Baidu cloud disk cuDNN compressed package extraction code: lsic
Replace tutorial.
Make sure the webui is closed and then open the compressed package, decompress and replace these files into the folder of the integrated package.
Feel the thrill of generating images in batches.
A total of 32 pictures took less than 30 seconds, so fast~
