Show HN: DynamiCrafter: Animating open-domain images with video diffusion priors

[Doubiiu]

https://github.com/Doubiiu/DynamiCrafter

Hello HN! I would like to show our recent research project: DynamiCrafter.

DynamiCrafter can animate open-domain still images based on text prompt by leveraging the pre-trained video diffusion priors. It supports:

- Image-to-video generation

- Storytelling video generation

- Looping video generation

- Generative frame interpolation

We have released source code at https://github.com/Doubiiu/DynamiCrafter and deployed a demo at https://huggingface.co/spaces/Doubiiu/DynamiCrafter

Please also refer to https://www.youtube.com/watch?v=PtW7hjCawbo and https://doubiiu.github.io/projects/DynamiCrafter/ for more results.

[Reubend]

Great results! This is the most consistent video output I've seen so far from a single frame.

[whycome]

Is open-domain the same thing as public domain?

[Doubiiu → whycome]

Ah, I'm not sure about 'public domain'. In this context, 'open-domain images' refers to any images, as opposed to curated domain images, such as human, fluid, cloud, etc., which other traditional image animation methods are specifically designed for.