Cannot repeat outputs from image_log

[Jakubelo]

Hi, I am having issue to get similar outputs as in image_log. I have trained the model on sd2 core with custom dataset. I also prepare gradio_foo.py script to make an inference and I writer another script on based on ControlLDM class(to use the same params as training) and image log method, but non of these approaches gave me similar results as results in image_log during the training. The outputs don't need to be the same, but quality of output have to. Almost all images from training are good enough for end solution, but inference on gradio is giving me completely unuseful outputs.

So the main question is, how to make outputs with similar quality?

Unfortunately I cannot show any examples, the data is private.

[Khaihuyennguyen]

Hi, I have the same problem, do you have any updates?

[ohcccc]

excuse me.i also have the same problem.do you solve this problem?

[Jakubelo]

@lllyasviel do you know maybe the reason? I followed every part of the code and mostly params look similar. I know that I have probably small dataset, but function log_images produces significantly better results than inference script that I made which follows other grade files.