CUDA OUT OF MEMMORY FOUND SOLUTION

[ChaosMonke]

You just have to edit web user to have set COMMANDLINE_ARGS=--medvram --precision full --no-half if it does not sitll work try using --lowvram instead

[amiketanner]

Anyone have some other suggestions for dealing with the 'CUDA out of memory' than this?
This incredibly slows down image generation. From less than 30 seconds with 6GB VRAM to 5.5 minutes without Aesthetic embedding. At least Aesthetic embedding can now work at ~8 minutes, but I think it is much slower than it could be considering the dramatic increase in time without Aesthetic embedding. There has to be some better memory management strategies.

[Kilvoctu]

Something seem to change somewhat recently in Web UI that's causing it to take more memory.

This is the argumants I use for GTX1080 8GB VRAM:
--xformers --opt-channelslast
A sample generation, batch size 1, no embeddings or anything
(Steps: 30, Sampler: Euler a, Size: 512x512, Time taken: 15.63s, Torch active/reserved: 2713/3182 MiB, Sys VRAM: 3919/8192 MiB (47.84%))