mnb3000/runpod-stable-video-diffusion repository overview

⁠Runpod Stable Video Diffusion

This is a container for Runpod (or actually any other system) with all the dependencies for stable video diffusion to run.

⁠Dockerfiles and start scripts code available in GitHub repo⁠

⁠Description

By default it automatically checks for a /workspace directory (default path for runpod's external volume mount, which can be changed with the $MODEL_MOUNTPOINT environment variable), downloads the weights for the model there, and runs the official streamlit demo on port 3000 by default (this can be changed with $PORT environment variable)

If you are using this image on your local machine you will need to mount a host directory to /workspace (or path specified in $MODEL_MOUNTPOINT) with at least 20 GB of free space (40 GB if downloading both models)

There are 3 different tags available: svd-base-xt downloads both models, and svd-base/svd-xt download only one model respectively

⁠Templates

You can use it as a template for a pod: BASE⁠, XT⁠

⁠Example run commands (replace /your/path with your directory)

⁠With default model path/port

docker pull mnb3000/runpod-stable-video-diffusion:svd-xt
docker run --gpus all \
    -p 3000:3000 \
    -v /your/path:/workspace \
    -t mnb3000/runpod-stable-video-diffusion:svd-xt

⁠With custom model path & port

docker pull mnb3000/runpod-stable-video-diffusion:svd-xt
docker run --gpus all \
    -p 8080:8080 \
    -e MODEL_MOUNTPOINT=/model \
    -e PORT=8080 \
    -v /your/path:/model \
    -t mnb3000/runpod-stable-video-diffusion:svd-xt