ImagineArt
Putting a generation engine everywhere the work already happens
- Role
- Product Engineer — surface strategy owner
- Where
- ImagineArt (Vyro) · Jun 2026 — Present
- Status
- Shipped, in production
People don't want to visit your agent. They want their agent to use your product, without leaving the tool they already have open.
millions
Users on the platform
5
Host apps shipped into
1
Browser extension released
What I own
ImagineArt reaches millions of users through its own apps. Professional creatives were not among them, because professionals do not leave the software they bill hours in to go generate an asset somewhere else and carry it back. The destination model quietly excludes them.
So the question I own is not "how do we get creatives to our product" but "how do we put the product where the work already happens" — which turns out to mean host apps, browsers, and agent runtimes, not one more web page.
Plugins
Five production surfaces — Premiere Pro, After Effects, Photoshop, Figma, and Framer — each a real host runtime with its own sandbox, lifecycle, and distribution story rather than a webview with a logo on it.
The Premiere Pro one runs on UXP with a pull-first architecture and ships as a notarized .pkg for macOS. Generate, upscale, clean up shots, and conform Film Studio projects into editable timelines without leaving the edit.
The browser extension
Released on the Chrome Web Store: AI face swap, recreate, and style grab on any page you browse, in a side panel wired to your ImagineArt account.
The pitch is the whole thesis in one line — the browser is the world's biggest mood board, so act on any image on the web without leaving the tab.
Agents and the MCP server
Alongside the surfaces I built an agentic video editor, and I maintain ImagineArt's MCP server — moving it toward a lean core tool surface with skills layered on top, so tool selection stops degrading as context grows.
That is the same problem I hit at 120+ tools on TradersGPT, arriving from the other direction.
What the runtimes teach you
None of it is documented. What the sandbox will not permit, which integrations look obvious and turn out to be impossible, which constraints only surface in week three once you have already committed to an approach — you learn it by shipping and getting rejected.
That accumulated knowledge is the actual asset. It is not a model problem, so the labs will not solve it, and it does not demo well, so most teams will not touch it.
What happened
These surfaces are now a distribution channel into professional users the company previously could not reach. That is company revenue rather than mine, and I would rather say so plainly than dress it up.
Built with