Open seandearnaley opened 2 years ago
Here is some more evidence of this issue, take a look at the screenshot below from colab-- this is not working as expected... the issue is the colab is fresh and therefore has no artifacts like prevFrame.png
, based on this code in disco.py
I would expect the script to pick up from the chosen batch and frame but instead it seems to set the img_path
to prevFrame.png
which of course doesn't exist... this pretty much prevents me from picking up from a timeout batch-- long colab sessions timeout:
unfortunately I'm not great with python or I'd try to fix this myself
if resume_run and frame_num == start_frame:
img_filepath = batchFolder+f"/{batch_name}({batchNum})_{start_frame-1:04}.png"
if turbo_mode and frame_num > turbo_preroll:
shutil.copyfile(img_filepath, 'oldFrameScaled.png')
else:
img_filepath = '/content/prevFrame.png' if is_colab else 'prevFrame.png'
Confirmed this issue doesn't affect disco diffusion 5.1
Im not sure what update broke this functionality but there has been a regression from disco diffusion 5.1- I seem to have lost the ability to resume a batch from a fresh instance, either via colab or running locally... if I try to pick up a batch from a previous frame I get the error file not found
prevFrame.png
, it doesn't really make sense because the previous frame in this case should be the init frame I've selected and as the instance has never run before, there will be no artifacts therefore no previous frame. This used to work fine for me, but now animation resuming doesn't work at all unless I'm in the same session and that is a bit of a showstopper considering timeouts/ I wonder if this is related to @aletts recents updates.