Question about the training details.

[zouxinyi0625]

Hi, thanks for your great work.
I try to run the code with the SD1.5 model, v3_sd15_mm.ckpt, and v3_sd15_adapter.ckpt with the default setting provided in default configs. But the generated result is not satisfactory as shown in your Readme. Is the demo trained by multiple videos, or something misaligned with the default config?
400 Steps training (w domain adapter):

[FredrikLaastad]

Wondering the same, as I am also completely unable to reproduce the results shown in the readme.
I am noticing that there seems to be no noticable difference in the samples produced between 150-500 steps. So something seems to be off.

100 steps

200 steps

500steps