Poll results: inpainting and text2video (Patreon)

Hi there! Looks like there were only two major suggestions in that 100-patron-goal poll: inpainting and video generation. 

Inspired by that, I've added inpainting into warpfusion - our video generation tool :D 

I will edit this post later to incorporate things I find on those two topics, including papers\code and colabs. 

As for now, no decent (IMHO) public open-sourced text2video is available, given the fact that it's one dimension harder than text2image generation, and we've just started getting cool-looking results this year. 

On the other hand, on-point inpainting tools are widely available.

Inpainting

LAMA
github

You can begin with LAMA, the one being used in #warpfusion
Their colab with interactive mask drawing is available here
Huggingface web demo 

text2video

CogVideo

github 

A well known text2video model from China.
I'd suggest using their web demo and huggingface space, because their recommended inference GPU is an A100 :D

HF space now has an init image as well

NUWA

github 

A swiss-army-knife of models by Microsoft. Offers all kinds of text-driven genertive and manipulative tasks, but sadly, no official pretrains or even code are available. 

There is a custom implementation by lucidrains (God bless his soul) here
Last commit in april, so it's a bit stale.