stabilityai

/

stable-video-diffusion-img2vid-xt

like 2.28k

Helllo,

I appreciate your putting this together. I do have an issue, though. I keep getting the OOM error within the Streamlit app. I have the Colab Pro plan, and have adjusted the frame rate, but still can't get anything to go through. I did try to adjust the video size, however, I received an error, which seemed to indicate that it needed to be the size you have it set at, so I adjusted it back. Here is the error:

OutOfMemoryError: CUDA out of memory. Tried to allocate 1.10 GiB (GPU 0; 15.77 GiB total capacity; 13.03 GiB already allocated; 294.12 MiB free; 15.09 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

Any help would be appreciated.

Thanks

What size of image are you running through? I haven't been able to run any large high definition images through on the V100s with 15GB of memory. I'd recommend putting it on an A100 if you want to animate a large image. One good thing to try is grab their test image and make sure everything is working by running it through https://github.com/Stability-AI/generative-models/blob/main/assets/test_image.png. I don't remember what the dimensions are but you can find them by opening the image in most editors or using Pillow. : ).

Can't we run it without a A100 GPU?

You can! For larger images you might run into some memory issues like above though :/.

I keep getting an error about the scripts module not being found, even though I followed your instructions in the beginning to do the sys.path.append().

Paths can be such a pain. What directory are you running streamlit from? Can we check which directory, and that the paths for sys.path.append are correct?

The command "! cd generative-models && streamlit run scripts/demo/video_sampling.py &>/content/logs.txt &" runs forever or sometimes it returns "Killed"