Captions

Captions is executing one of the cleaner consumer-to-lab transitions in AI: the app that more than 20 million creators use to automate short-form editing (captions, filler-word removal, eye-contact correction) became the data and distribution engine for Mirage — the September 2025 corporate rebrand signaling a foundation-model company purpose-built for short-form video, with models trained specifically on pacing, framing, and attention dynamics rather than generic video generation. The March 2026 $75M growth financing from General Catalyst's Customer Value Fund took total funding past $175M (atop the 2023 Kleiner Perkins-led Series C that valued the company around $500M) and, notably, funds customer acquisition against proven customer value — a structure only available to companies whose unit economics already work. The AI Actor studio is the strategic sharp end: fully synthetic performers generated from a prompt, aimed at the advertising and marketing spend that pays real money for scalable spokesperson video, and a natural bridge from consumer subscriptions into B2B revenue. The compounding loop is the thesis: millions of creators generate engagement data on what makes short-form video perform; that data trains models no lab without a consumer surface can replicate; better models improve both the consumer app and the ad-facing products; and planned expansion into agentic video editing and Asian markets widens the funnel. In a world where short-form video is the dominant media format and its production is being automated end-to-end, Mirage owns a defensible position at the intersection of consumer scale, proprietary training signal, and vertical model focus.

Mirage has chosen to fight a foundation-model war against opponents with three orders of magnitude more compute. OpenAI's Sora and Google's Veo generate increasingly convincing human performances from prompts as by-products of general video models; ByteDance's CapCut owns short-form editing distribution through TikTok itself and ships AI features to a user base vastly larger than Captions'; and Adobe, Meta, and TikTok's Symphony all press into AI ad creative. A vertical model "specialized for short-form pacing and attention" is a real idea, but history suggests frontier general models absorb vertical advantages within a generation or two — and when they do, Mirage's differentiation reverts to app-layer UX, where CapCut's distribution wins. The financing structure deserves scrutiny rather than celebration: $75M from General Catalyst's Customer Value Fund is not a priced equity round — no new valuation mark has been disclosed since the $500M of mid-2023 — and CVF capital specifically finances marketing spend against future customer revenue, a signal that growth requires paid acquisition and that a conventional up-round may not have been the preferred (or available) path. The consumer creator-subscription base that funds everything is notoriously churn-prone and increasingly courted by free, platform-native tools. The AI-actor bet also walks into the teeth of synthetic-media regulation and brand-safety caution just as fully synthetic performers face disclosure mandates. Mirage must out-model giants, out-distribute CapCut, and convert consumer scale into enterprise ad revenue simultaneously — three hard races, one balance sheet measured in hundreds of millions, not billions.