yanwu xu
yanwu xu
still waiting for google to provie, stay tuned.
Hi, I update the code for simulation for now.
I wrote an unofficial implementation of UFO on the DiT imagenet, but the behavior is different from the text2image generation, which only supports 4 steps of generation.
Hi, sorry for the long-term no-reply. We only test it on the T2I models, to get one-step, you can try to reduce the gap between the xt and xt-1.