Project Overview

Paper2Video takes in a paper converts it to an educational video that explains it, using ChatGPT and Dall-E.

The way it works is it reads the chunks of the paper (due to context limits) and generates a script for the video, including prerequisite concepts. Then each line gets converted to a description of an image, and each description gets some modifiers like “oil painting of” and “ultra HD”. That is then sent to DALL-E, and the script text is sent to a free text to speech api (which is why it’s so bad).

Examples

I recommend watching at least 1.5x speed due to the voices.