my generated video is black

[lu123456789]

Why is the generated video black and the value of sample is nan?

[hcl14]

Same for me. I fixed this by cloning from huggingface then cloning the same stable diffusion 1.4 as gradio demo wants git lfs clone https://huggingface.co/CompVis/stable-diffusion-v1-4 into checkpoints/CompVis/stable-diffusion-v1-4 (this path is hardcoded in some places). Then I can generate videos by editing configs to point into this path, also you need to set use_null_inv: false in config and probably mixed_precision: fp16 otherwise you get bad-looking result. Gradio demo is broken and can't find the model it downloaded (perhaps hardcoded path), I used test_vid2vid_zero.py.

You may have error Error: mkl-service + Intel(R) MKL: MKL_THREADING_LAYER=INTEL is incompatible with libgomp.so.1 library - you need to reinstall numpy

[lu123456789]

Same for me. I fixed this by cloning from huggingface then cloning the same stable diffusion 1.4 as gradio demo wants git lfs clone https://huggingface.co/CompVis/stable-diffusion-v1-4 into checkpoints/CompVis/stable-diffusion-v1-4 (this path is hardcoded in some places). Then I can generate videos by editing configs to point into this path, also you need to set use_null_inv: false in config and probably mixed_precision: fp16 otherwise you get bad-looking result. Gradio demo is broken and can't find the model it downloaded (perhaps hardcoded path), I used test_vid2vid_zero.py.

You may have error Error: mkl-service + Intel(R) MKL: MKL_THREADING_LAYER=INTEL is incompatible with libgomp.so.1 library - you need to reinstall numpy

Thank you very much. It's very helpful and useful.