Mass Produced AI Video

Go from 1 sentence prompts to voiced & animated YouTube videos.

Comment

Inspiration

I was inspired by YouTube videos like the Balenciaga and Wes Anderson homage videos which were made by AI and described in this recent article.

https://thechainsaw.com/nft/heres-how-to-make-those-viral-ai-balenciaga-harry-potter-ads-without-much-effort-at-all/

I especially like YouTube as an application for this because you can create as much content as you want, so not only as it easier to 'rank' and attract an audience than with, say, SEO, but you have as many opportunities to try to build an audience - at no cost - as you have the time to attempt. Because of the historical difficulty level of creating video, I'd say you also have less competition at it.

What it does

You give the app a 1-sentence description of what to do and it elaborates that into 10 bullet points, which are then illustrated (3 images generated now) and accompanied by voice narration of those bullet points. This is meant to emulate, roughly, what an automated YouTube content strategy would look like.

How we built it

I built it using React, the Open AI API for text & image generation, and the AWS Polly service for voice generation.

Challenges we ran into

The challenges, which I didn't see coming, were: rate-limiting, which does put a cap on how quickly you can move through the process (at least when using Open AI API), and activating back end services like ffmpeg, which basically was too time-consuming to be included. Instead video was 'simulated' with an endlessly repeating gif and audio playing over it.

Accomplishments that we're proud of

This was a lot of services to hook together and effectively use to create something useful in ~2 days.

What we learned

If this much is possible within 2 days, then great things could be accomplished within a month. Also news to me to find out that, w/AI as an accelerant, 2 days is now enough to produce a real prototype.

What's next for Mass Produced AI Video

Improving every element of it - the voices (for example: it would be easy to have dueling voices), removing the speed limits (figuring out a faster method to generate images), while balancing with important considerations like cost - and finally seeing what kind of elements bring popularity on YouTube, once videos start going up there.

Built With

Share this project:

Updates