giraffe.study

giraffe.study is revolutionizing the way we learn by creating customized video content for your needs as a learner, a creator enhancing production, or a researcher distilling complex papers.

Comment

Inspiration

One of the best ways to learn about a new and complex topic is through a good video explainer. After all, video has the highest bandwidth of communication, involving visual, textual and auditory information to aid in understanding. One example of this is a channel like 3Blue1Brown, a channel familiar to many of us. However, it can take weeks for creators like 3Blue1Brown, to research, script and animate a single video, meaning such videos are actually in short supply, with especially low coverage for advanced academic topics. We recognized this gap and felt inspired to bridge it by building an AI video generation tool that can both autonomously generate explainer videos on any topic, and boost efficiency for existing creators.

What it does

Imagine a world where any academic or conceptual query is met with a detailed, visually rich video explanation—almost instantly. That's the world giraffe.study aims to build. Our tool is perfect for:

  1. Learners: Get on-demand, AI-generated video explanations in the style of 3Blue1Brown.
  2. Content Creators: Boost your productivity! Create content at tenfold speed, ensuring consistency and quality.
  3. Researchers: Ever been daunted by a research paper's complexity? Just upload the PDF, and let our tool generate a concise video summary for you.

Our tool is designed for learners to ask about any topic and get an AI-generated video-explainer in 3Blue1Brown style; for content creators to streamline and speed up their content creation process, and for researchers to be able to upload a pdf of a paper and get a quick video explanation of it.

How we built it

At the current time AI generated video is at the very frontier of what’s possible, and it’s currently impossible to use AI to produce a coherent video longer than a minute. However, we managed to harness recent advances in code-generation models such as GPT4 and open source libraries such as manim to make coherent video explanations possible, by generating videos through code rather than as images. To achieve this, we create an agent using Open AI API that autonomously plans a lesson, writes code to produce visualizations, and catches its own mistakes. Our process starts with topic input, which GPT-4 uses to generate a detailed script. This script then informs the creation of individual video segments, which are seamlessly stitched together. The finishing touch? A lifelike narration courtesy of the text-to-speech API from Eleven Labs.

Challenges we ran into

It was hard to link all of the files to the backend and a little tricky to manage this big and modular project for a small team. However, we got everything to work!

Accomplishments that we're proud of

There are currently no tools on the market that offer AI-generated videos longer than 1 min. It takes a human about a week to create a 3-min video with 3Blue1Brown style, while it takes us around 5 min. Yes, the quality of our app is not as good as the human-generated videos, but we can do it ~90x faster. Additionally, our team has worked super smoothly and collaboratively on managing this big project. The journey from a conceptual Figma prototype to a fully functional platform was exhilarating. We're particularly proud of the behind-the-scenes magic: the files that craft video scripts, create individual video segments, add narration, and bring it all together into a cohesive video—without the user ever sensing the complexity.

What we learned

Firstly, handling a project of this magnitude taught us the importance of modular design, effective communication, and the art of smooth software deployment. From the technical perspective, we have learned to use Node.js to write backend, create a localhost, parallelize code and make it more efficient, manage bigger modular projects, use API from OpenAI and from ElevenLabs, write Bash scripts, smoothly streamline software development and deployment.

What's next for giraffe.study

Firstly, we would like to improve on the code that we wrote in 36 hours, some of which we spent sleeping, eating, and relaxing, as the horizon is vast for giraffe.study. We're excited about the prospect of adding multilingual support, making knowledge accessible across linguistic barriers. To support more learning styles and offer diverse video styles for everyone, we envision collaborations with renowned educational content creators. Additionally, with the integration of academic databases, giraffe.study could become the go-to platform for video summaries of the latest research breakthroughs. We're not just creating videos; we're crafting the future of education.

Share this project:

Updates