Regarding the consistency of images generated multiple times, because the algorithm uses a diffusion model, can the details of images generated multiple times maintain consistency for a single target pose?
The pretrained model is a 512512 stable diffusion model, would you please tell me about how to generailze it to 512352 or 256176 even 12864 in market-1501? I tried to pass the height and width in the pipeline but produced messy up images.