This the Gradio variation of https://github.com/nv-tlabs/GEN3C to run on https://modal.com/. Currently Running on a Single A100-80GB GPU with inference time for each clip around ~15 mins. Modal offers 30 dollars for free for GPU computing which is more than enough to run this model for couple of inferences!
Make sure you have Modal module installed You have to always make sure that you have requested access to https://huggingface.co/nvidia/Cosmos-Tokenize1-CV8x8x8-720p repo as it's a gated repo
python3 -m pip install Modal
and also set up correctly
python3 -m modal setup
you would also need a huggingface Token set a Secret with the name HF_TOKEN on modal dashboard then simply deploy the model
git clone https://github.com/OutofAi/GEN3C-MODAL
cd GEN3C-MODAL
python3 -m modal deploy modal_cli.py
After deployment you will be provided with a link that's going to look something like this
https://{username}--gen3c-ui.modal.run
navigate to that page to find you Gradio setup
The first time runnning the Gradio interface, you would need to dowload the models on Modal's storage by clicking Download Checkpoints before running inference, it's 70+ GB of data and then after you wouldn't need to download it anymore as it gets stored on the given volume
