search

fofr / cog-consistent-character

Create images of a given character in different poses
https://replicate.com/fofr/consistent-character
MIT License
459 stars 40 forks source link

12 GB of VRAM, not enough? #9

Open divid3byzer0 opened 2 weeks ago

divid3byzer0 commented 2 weeks ago

I am trying to run the model on Docker (Docker Desktop, Windows via WSL2) and my card is a RTX 4070 12GB, but I always see the error "torch.cuda.OutOfMemoryError: Allocation on device" and the predictions although they say "suceeeded", there are no output files.

I am guessing that the minimum for this model is 16 GB of VRAM?

ArianaStar commented 1 week ago

@divid3byzer0 I noticed that this issue is a week old & is still open. Allow me to solve it for you.

Simply run this: python main.py --listen 0.0.0.0 --lowvram --preview-method auto --use-split-cross-attention Instead of this: python main.py --listen 0.0.0.0

If you get "Killed" just before the end of the process... Use this: --novram Instead of this: --lowvram

Hopefully I solved your problem, that solved it for me! enjoy. ♥