A look at Decart, which has developed video-to-video AI model Mirage that manipulates live video in real time and can potentially shake up livestreaming
A startup called Decart has developed an AI model that can transform live footage. The results are mind-bending—and poised to take over streaming.
Context & Ripple Effects
Mirage places Decart at the intersection of real-time generative video and the company’s GPU-optimization work for cloud providers and AI companies. The product angle matters because it moves video generation from a produced asset toward a live layer applied to an existing stream.
The later $100M Series B for Decart’s real-time AI and infrastructure work and subsequent $300M raise at an approximately $4B valuation show that investors treated this capability as part of a broader real-time AI platform, not solely a consumer video feature.
First-order effects
- Livestream creators and production teams can potentially alter live footage as it is broadcast, shifting generative video from post-production into the live workflow.
- Decart must deliver low-latency, reliable transformation at streaming scale; its GPU-optimization technology becomes directly relevant to the product’s usability and cost.
Second-order effects
- Streaming platforms and live-production software providers face pressure to offer comparable real-time creative tooling or integrate third-party models, while distinguishing authentic footage from transformed output.
- As live synthetic video becomes easier to deploy, provenance tools gain importance; Meta’s earlier video watermarking and origin-detection tool illustrates the adjacent need to identify AI-generated material.
Third-order effects
- If real-time transformation becomes dependable and widely distributed, livestreaming could shift from a format centered on captured footage to one where visual presentation is dynamically generated, raising the value of compute efficiency and distribution access.
- The same capability makes disclosure, provenance, and moderation harder in time-sensitive settings; market adoption is likely to increase demand for control systems rather than eliminate the need for human platform governance.
The trend: This is one data point in the shift from offline generative-video creation to real-time, infrastructure-intensive synthetic media embedded in live workflows.