Hey all, wanted to share a first draft of what I’ve been working on!
Aeon takes written documents, such as web pages, Google docs, even audio transcriptions, and turns them into high-quality ‘explainer’ style videos. The videos have a professional look and feel, with graphics and structure rivaling that of what human producers do.
In this example, we took the following, admittedly very boring website:
Check out the walkthrough here:
The user just gives very few instructions: Make a commercial, focus on the partners at the firm, etc.
Then a whole bunch of AI tools take over: GPT3, BLIP, NLP tools, Elevenlabs TTS, facial detection, etc. There are still some glitchy bits, but I’ll iron those out.
What was super fun (and challenging) was that GPT3 actually “directed” the visuals. I have approx a dozen scenes and styles that I describe to GPT3, which then decides what type of visual and motion graphics would work for each scene.
For the Shotstack folks: thank you for making a great product! I definitely have suggestions as to what would make my life easier. Would love any and all feedback you all have too.
3 Likes
This is amazing! I love it.
Would love to hear more about your experience building this.
Will you be looking to make this available as a product?
2 Likes
Yes! Just working on some more use cases.
1 Like
Lookin’ good drfalken!
Since, I’ve been working on a similar project, it’s really cool to see your take on this process. I can tell you’ve put a lot of work into it.
Well done.
Thank you! I’ve got a site up: https://project-aeon.com/ and a twitter feed: https://twitter.com/get_aeon
Would love to see what you are up to as well!
Your project is at least 3 months ahead of mine, so I’m afraid I mostly only have proof-of-concept artifacts laying around in terms of show-and-tell.
I AM, however, just starting an open-source project to make an alternate template format for shotstack. The intention is that it can convert from it’s format into shotstack’s format, so it could be used as a webservice/command-line/library to preprocesses feature-rich templates into shotstack templates.
Features such as…
- Z-index instead of tracks
- Scenes and other modular templating techniques like partials
- Variables, Scopes, and evaluated expressions
- Optional and standardized units of measure
- Lazy loading
Repo is empty as of this post, but it’s here, FWIW: GitHub - CobaltBlueDW/ShotstackElements: A Template Framework for Shotstack
It’ll probably have something substantial to show in a few days.
1 Like
Very cool, looking forward to it!