tin2tin/Pallaidium

Problem: Video, img2img

b4zz4 opened this issue · 4 comments

b4zz4 commented
Seed: 747642197
Process: Frame by frame (SD XL)
1/5
100%|███████████████████████████████████████████| 12/12 [00:28<00:00,  2.39s/it]
2/5
100%|███████████████████████████████████████████| 12/12 [00:30<00:00,  2.56s/it]
3/5
100%|███████████████████████████████████████████| 12/12 [00:30<00:00,  2.50s/it]
4/5
100%|███████████████████████████████████████████| 12/12 [00:28<00:00,  2.41s/it]
5/5
100%|███████████████████████████████████████████| 12/12 [00:28<00:00,  2.41s/it]
Traceback (most recent call last):
  File "/home/bazza/.config/blender/4.0/scripts/addons/Pallaidium-main/__init__.py", line 2129, in execute
    video_frames = np.array(video_frames)
                   ^^^^^^^^^^^^^^^^^^^^^^
ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detected shape was (5,) + inhomogeneous part.

Can you upload a screenshot of your generation settings? And the dimensions of the source video/image?

b4zz4 commented

I test with 1024x512 and 512x512 for input image and output video

It has problems with the Quality Step at the value of 1. So, now it is limited at 2.

This img2img model is not super useful because there is no temporal consistency, and things ends up flickering a lot. I might disable this option at some point.

SD Turbo needs a Quality Step value of 1, so I can't cap it a 2 for img2img, so I have disabled img2img for now.