We finally cracked it!
We built an open-source video editor that turns every edit into code.
I got tired of agents:
→ Losing context
→ Giving me outputs I couldn’t modify
So we flipped it:
→ Code is the new database.
I gave Astra an old youtube tutorial + access to a Figma landing page. 20 mins later I had this to work with for some tweaking, inside your full blown editor. Truly amazing.
If you need some marketing support, DM is open ;)
In Diffusion Studio we switched from tracks → layers for exactly this reason.
In our experience, agents do a lot better when they can group, nest, and move things around in a scene graph, similar to how they work with the DOM.
Is NLE the right approach? It’s an abstraction framework to help humans understand video editing.
The tracks don’t actually exist and it’s exactly where ai fails with raw composition.
Agents perform much better with layer based composition (like ae) than with NLE/tracks.
Figma for video editing doesn't exist for a reason.
At my company, we're building a video editor for AI. But I've been eyeing this space for years.
"Figma for X" has been a popular startup pitch since 2019. And serious founders have certainly tried, with millions in funding, to
Video editing will change forever when people start sharing their edits on GitHub.
I want a future where someone asks "how did you make this?" and you just send them a link to the project source that agents can also read and understand.
The brief, project source, timeline,