Make graphical animations with GPT-4

[scenic jacinth]

I've built a simple prototype to generate animations with GPT-4 using Manim:

How it works?: https://www.youtube.com/watch?v=zjh_7cg6Aa4
Prototype App: https://generative-manim.streamlit.app
Repository: https://github.com/360macky/generative-manim

Manim is a Python library for creating complex animations. It's the library used by channels like 3Blue1Brown and many more.

Since GPT-4 is getting better at programming. I was wondering what would happen if we connect the two. I call it Generative Manim.

This is just a quick experiment to test GPT-4 in animations. An improved production-ready version would have many more things: a continuous error correction for certain parts, maybe a real-time editor, etc.

[languid token]

This is cool; I wonder if it would be possible to upload an image and somehow makes generated animations based on that image?

[scenic jacinth]

That sounds really good!
I’m not sure how can we achieve that with current technologies. What do you think?

I was thinking of GPT-4 Multimodal could make this idea reality: Request the Manim code of an image provided by user, then get the code and modify it to enable motion and finally render the video.