The VAE from openai is quite good when structural objects in images such as lines, texts, etc. But it really costs a lot of VRAM comparing to AutoEncoderKL.
· Sign up or log in to comment
