Safetensors

Make several specific models

#1
by Fabrice-TIERCELIN - opened

Hi @Artiprocher ,

You say that 40,000 videos take a week. What about creating a model with only drone videos to only create remote camera motion? Then you create another model with static cameras only. Then a model trained on fire videos, water videos, people videos, dog videos etc...

  1. Each model will be trained quickly
  2. Each model will be more realistic
  3. You will publish models regulary
  4. You can suggest all the models in one space like here
  5. The space would become almost multimodal (image + video type choice)
Computational Intelligence Lab at ECNU org

This is an interesting idea. However, I think it's more important to train an all-in-one model. We focus on improving the general abilities of base models.

OK. And do you know custom models on the web for Stable Video Diffusion other than "stabilityai/stable-video-diffusion-img2vid" or "stabilityai/stable-video-diffusion-img2vid-xt-1-1"?

It's for completing options for spaces like xi0v/Stable-Video-Diffusion-Img2Vid

Sign up or log in to comment