Interview

Decart founder on Mirage: real-time AI that applies 'cyberpunk' and other styles to live video streams for free

Jul 17, 2025 with Dean Leitersdorf

Key Points

  • Decart launched Mirage on July 17, 2025, a real-time video model that applies user-defined visual styles like 'cyberpunk' or 'Lego' to live camera feeds with server-side processing.
  • The team optimized GPU assembly code to achieve inference costs low enough to offer the product free, targeting ad-based or platform monetization instead of subscriptions.
  • Live streaming monetization through tip-triggered style changes is the clearest near-term revenue path, extending existing green-screen tools into full-scene aesthetic transformation for creators and viewers.
Decart founder on Mirage: real-time AI that applies 'cyberpunk' and other styles to live video streams for free

Summary

Decart, a research lab less than two years old, launched Mirage on July 17, 2025, a real-time video transformation model that applies user-defined visual styles to any live camera feed or video stream. The product went live roughly 30 minutes before the segment aired, making the interview its first public demonstration on a video call.

The core capability is prompt-driven style transfer applied in real time. Users type a descriptor, "Lego," "cyberpunk," "cosmic medieval golden," "wizard," and the model instantly re-renders the live video feed in that aesthetic. The system is not running locally; it processes webcam input server-side and returns the transformed stream with low enough latency to be usable in live conversation.

Founder Dean says the team wrote low-level assembly code for GPUs to achieve both the real-time performance and the unit economics required to offer the product free to consumers. The inference cost is low enough, he argues, to support ad-based or platform monetization without subscriptions, making it viable as a consumer product layer rather than a premium tool.

The clearest near-term commercial channel is live streaming. The segment discussion converged on a tip-triggered interaction model where viewers pay to push a streamer into a chosen visual style, slotting Mirage into existing streaming monetization mechanics. Streamers already use green screens and background replacement tools, and Mirage extends that pipeline to full-scene style transformation. Creator-side use cases were also floated, including a children's content creator offering a "Lego version" of a video as an alternative viewing mode.

Enterprise adoption faces an obvious cultural barrier. Showing up to a Fortune 500 Zoom call as a Lego figure is the working example of where the technology likely will not land in the near term. The more plausible professional angle is client-side filtering, where an individual transforms their own view of a call without altering what others see.

Decart is positioning Mirage as a new category of consumer video interaction distinct from generative video platforms like Sora or Runway, which operate on pre-recorded or synthesized clips. The real-time constraint is the differentiator, and the company claims it is the only model currently operating at that latency threshold.