Most “AI for creators” tools are a thin web page over a server. Every edit, every drag, every keystroke is a round trip to someone else’s machine. The tool feels like a remote control, not an instrument.

We split it differently, and the split is worth explaining honestly.

Where the line falls

There’s exactly one thing in StoryVis that genuinely needs a giant model: generation. Turning a prompt and a handful of reference images into a frame is what hosted image and video models are for, and we use them. We’re not pretending the GPU in your laptop renders the film.

Everything else runs in your browser — compiled from Rust to WebAssembly:

  • the canvas you arrange shots on,
  • the image editor you fix a frame in,
  • the story document that holds your characters, scenes, and shots,
  • the orchestration that decides what to generate, in what order, and checks the result.

The cloud does generation. The studio is local.

Why that’s worth the trouble

The canvas is instant. Arranging shots, scrubbing, editing — none of it waits on a server. The interactions that should feel like a tool feel like a tool.

Your project is local-first. The story document lives in the browser as a CRDT. It works offline, syncs when you want it to, and merges cleanly when two people edit the same scene at once — no “someone else has this open” locks.

The editor is on-device. Color, masking, 32-bit float pipelines run on your GPU through WebGPU. You don’t upload a frame to adjust its grade.

So the honest version of “your work stays with you”: the project — your structure, your edits, your arrangement — never has to leave the browser. The generation step does send a prompt and reference images to a model provider, because that’s where the big models live. We’d rather draw that line clearly than blur it in a privacy promise we can’t keep.

The tradeoff

This is harder to build than a server with a web UI. Rust to WASM, a GPU story across three browser engines, CRDT merge logic, an orchestration layer that runs the same in a tab as it would on a backend. You take on real engineering to give up the round trip.

That tradeoff is the part we signed up for. The next posts will be the technical ones.