Framefields
The "code-first" angle is what hooks me. Video editing has been a visual discipline forever, but once you treat a project as code, everything changes. You get version control, reproducibility, and the ability to diff a generation script across iterations. That's exactly the workflow AI-native video needs, where a single prompt tweak should re-render the whole thing.
Open-source is the smart call here. The AGI-era video space moves fast, and a proprietary tool risks getting outpaced by community-built integrations. Releasing it lets researchers and studios bolt on their own model hooks without waiting for a roadmap.
I want to see what it looks like in practice. Is it a Python library, a CLI, or a visual editor with a code backend? The description is thin, so I'm holding my enthusiasm until I see a basic example. The idea is right, and I'm curious how the team executes it.