Mirage raised $75M without giving up a single share — and it's hiring ML engineers at $275K to prove the model works

The $75M Growth Round and What General Catalyst's CVF Model Signals

General Catalyst's Customer Value Fund just wrote its first generative-video check, $75 million to Mirage, the New York company formerly known as Captions, and the structure of the deal says more about the state of AI video than the number itself. The round is entirely non-dilutive: a revenue-share instrument, not an equity check. That distinction separates companies with compelling pitch decks from companies with actual unit economics.

CVF exists to fund customer acquisition for startups that have already proven they can spend a dollar and earn back more than a dollar. Pranav Singhvi, managing director of the CVF fund, told TechCrunch that Mirage's business equation is "extremely figured out," adding: "They know exactly how to spend that dollar and generate a very attractive ROI." General Catalyst's "Unbundling of Growth Equity" outlines the fund's core argument: funding customer acquisition costs with dilutive equity is an inefficient allocation of capital. CVF offers an alternative: non-dilutive capital specifically for scaling proven acquisition loops.

This is the same vehicle that put $1 billion into Grammarly. Mirage's check is smaller, but the signal is the same: General Catalyst sees a company that has moved past the "will people pay for this?" phase and into the "pour fuel on the fire" phase.

What non-dilutive growth capital demands

Traditional VC rounds trade equity for capital to spend on whatever the company needs, R&D, hiring, infrastructure, marketing. CVF capital is narrower. It funds the growth engine, and the growth engine only. That means Mirage isn't using this money to figure out product-market fit or build a go-to-market strategy from scratch. It's scaling acquisition in markets where the math already works, particularly high-growth Asian markets, where the company says demand has been extraordinary.

The company's own blog post makes the logic explicit: "AI has made building dramatically faster and easier. Features that used to take months now take weeks. Quality is still paramount, but it's no longer enough." CEO Gaurav Misra echoed that in the Fortune press release, arguing that efficient capital allocation and scaled distribution are the new defensibility. When the CEO of an AI video company says product quality alone won't win, and then raises non-dilutive capital specifically to scale distribution, the strategy writes itself.

The funding brings Mirage's total to more than $175 million. Earlier investors include Index Ventures, Kleiner Perkins, Andreessen Horowitz, Sequoia Capital, HubSpot Ventures, and Adobe Ventures, a syndicate that typically writes equity checks at earlier stages. The shift to CVF for this round tells you those investors likely still own their stakes undiluted, and General Catalyst was confident enough in the revenue mechanics to bet on returns without taking more ownership.

Why the structure matters for AI-video hiring

Non-dilutive growth capital tied to revenue performance pressures companies to hire for output, not research runway. Mirage isn't staffing up to explore. It's staffing up to ship and sell. Zero G Talent's board currently lists open roles at Mirage added in the past week, including ML Engineer, Generative Video and ML Engineer, Agentic Systems, both in Union Square at $175,000–$275,000. Those are production hires, not research fellows.

Singhvi's assessment was blunt: "Regardless of what the other tools are out there, Mirage is clearly ahead of the pack from a unit economics standpoint." In a market crowded with AI video platforms (Canva, D-ID, HeyGen, Webflow, Avataar), the investor placing the bet is explicitly saying the moat isn't the model. It's the efficiency of the machine that sells the model.

20 Million Users and the Captions Platform as Enterprise Beachhead

Mirage now counts more than 20 million users. Those users have produced more than 250 million videos. In the past 12 months alone, the Captions app logged more than 3.2 million downloads and generated $28.4 million in in-app revenue, per Appfigures data reported by multiple outlets. The company's March 24, 2026 press release put the user figure in the same range and added HubSpot, CoreWeave, and King to its named enterprise client list.

The number that matters most for the General Catalyst investment thesis, though, is this: 75 percent of Mirage's revenue comes from outside the United States. That international concentration is no side effect. It's the product of a creator-first go-to-market that scaled globally before the company ever built a dedicated enterprise sales motion. The Captions app, originally an AI captioning tool, found its earliest traction among individual creators and small businesses producing short-form video for TikTok, Reels, and YouTube Shorts. That base gave Mirage a data flywheel, millions of real-world editing decisions feeding model training, that most enterprise AI startups have to simulate or buy.

The freemium shift in January 2025 accelerated the dynamic. Mirage dropped its paywall to compete directly with ByteDance's CapCut and Meta's Edits, both of which were using free distribution as a growth weapon. The user base expanded. Revenue kept growing. That combination caught General Catalyst's attention.

The enterprise layer is being built on top of this consumer foundation, not in parallel to it. Mirage Studio, launched in June 2025, targets brands and advertisers with a text-to-video platform that generates photorealistic AI avatars from a script or audio file. The brand owns the avatar as IP. Pricing sits at $399 per month for 8,000 credits. The web-based marketing suite, designed for companies producing video in bulk, addresses a different workflow than the mobile app. Mirage plans to merge the mobile and web platforms into a unified product, removing the friction for businesses that currently toggle between two separate tools.

The dual-tier model, consumer subscriptions starting at $25 per month and enterprise at $399, gives Mirage a revenue base that is unusual for a company at its stage. The consumer flywheel funds foundation model research. The research produces capabilities that feed back into both the consumer app and the enterprise studio. CEO Gaurav Misra has described the structure explicitly: the company is not building a feature layer on top of someone else's infrastructure. It's building the infrastructure itself, and using the consumer platform as the distribution and data engine that makes the research viable.

For the AI-video workforce, the implication is concrete. Mirage's 20-million-user base is not a vanity metric. It's the reason the company can justify hiring ML engineers for agentic systems and generative video at six-figure salaries while still meeting the unit-economics thresholds the fund requires. The enterprise beachhead is forming because the consumer product already proved the model works at scale.

The Generative Video Talent War: Who Mirage Is Hiring and Why It Matters

Mirage's hiring board tells a story that press releases don't. The company has 142 employees listed on LinkedIn, though third-party estimates range from 136 to roughly 557, a spread that likely reflects timing differences and whether Captions, its affiliated consumer app, is counted in the total. What matters more than the exact headcount is the composition: the open roles reveal what a generative-video company building enterprise infrastructure actually needs on the ground right now.

The core of the current hiring push sits at the intersection of ML research and agentic systems. Three of the most recently posted roles, all listed at $175,000 to $275,000 annually, all based in New York, are ML Engineer, Generative Video, ML Engineer, Agentic Systems, and Software Engineer, Agents. That pairing is the signal. Mirage isn't just staffing up video modelers; it's building the orchestration layer that wraps those models into production workflows. The "Agents" and "Agentic Systems" titles point to autonomous video-editing pipelines, systems that interpret natural language instructions and execute multi-step editing decisions without a human driving each cut.

The rest of the open roster fills out the platform around that core. A Software Engineer, Backend and a Software Engineer, iOS (both at the same salary band) suggest Mirage is simultaneously scaling its API infrastructure and maintaining a consumer-facing mobile presence. On the go-to-market side, two Senior Performance Marketing Manager roles ($150,000–$190,000) and an Influencer Marketing Associate ($70,000–$90,000) indicate the company is monetizing through both creator-economy channels and direct enterprise sales, a dual motion that matches its consumer scale and its B2B pitch.

The salary bands themselves are worth noting. Base compensation of $175,000 to $275,000 for ML roles in generative video puts Mirage in direct competition with OpenAI, Runway, and the well-funded foundation-model labs for the same talent pool. Zero G Talent's own board shows OpenAI posting agent and model roles in the $295,000–$445,000 range, while Runway's non-engineering roles sit lower, around $100,000–$210,000. Mirage is pricing its ML talent at the upper end of that market, a bet that experienced generative-video engineers are worth the premium because they can build proprietary production systems, not just fine-tune existing models.

The geographic concentration is just as telling. Every technical role lists New York. For a company operating in a talent market where remote-first hiring is the default for most AI startups, that clustering is a deliberate choice, and it means Mirage is drawing from, and contributing to, a growing local pipeline of generative-media engineers.

Foundation Models for Video: The Technical Moat Behind the Funding

The $75M growth round closes in a market where the underlying model layer is splintering fast. OpenAI's Sora, Runway's Gen-3 and Gen-4, Pika 2.1, and other generators are all competing on raw generation quality (photorealism, motion coherence, prompt adherence), and the gap between leaders and the pack is narrowing by the quarter. When the commodity layer improves that quickly, the moat isn't the model. It's what sits on top of it.

This is where Mirage's hiring pattern tells the real story. The company isn't just staffing a research lab or a product team. It's building the connective tissue between a model that generates video and a system that can orchestrate, edit, revise, and ship that video inside a real workflow. That distinction matters for enterprise buyers who don't need another text-to-video demo. They need footage that lands in a Premiere Pro timeline, a CapCut project, or a brand-approved asset library without a human re-cutting it from scratch.

The 2026 generative video landscape has crossed what Google DeepMind and OpenAI both flagged in 2025: video generation is becoming infrastructure, not a feature. A workshop on generative foundation models at CVPR 2026 laid out the technical frontier (long-form video via Sora and Runway Gen-3, production-ready 3D, multimodal reasoning in GPT-4o and Gemini), but the enterprise production stack still lacks a layer that ties these outputs together. The companies that win the next phase won't be the ones with the prettiest demo reel. They'll be the ones whose models plug into post-production tools natively and whose agentic systems handle the repetitive production work, trimming, formatting, versioning, that currently eats a creative team's week.

Runway, by contrast, is currently hiring for communications, social media, tax, and GRC roles, the profile of a company investing in go-to-market and compliance infrastructure rather than deep model R&D. That's a valid strategy, but it reveals a different theory of the game: Runway is commercializing its existing generation tech. Mirage, with dedicated generative-video and agentic-systems ML roles sitting side by side on its job listing, is betting the next competitive edge is vertical integration between the model and the production agent that wrangles it.

For anyone tracking where generative-AI-video engineering jobs are actually forming, the signal is clear: the hiring is concentrating at the model-plus-agent layer, not at the point-and-generate tool layer. Zero G Talent's board updates directly from company ATSes, and Mirage's current open roles are live alongside comparable listings at OpenAI and Runway.

Does Geography Still Matter for Generative Video?

Mirage sits in Union Square, New York, a short walk from a cluster of generative-media startups that have turned the city into a dense AI-video hiring market outside San Francisco. The concentration matters because generative video is not a fully remote-first talent category: the engineers who can train and optimize video foundation models tend to cluster where the production clients and the post-production infrastructure already exist.

Built In lists 61 generative-AI startups in New York. Exploding Topics tracks 62 AI startups in the city with enough search volume to rank. The F6S database shows 41 generative-AI companies and 100 AI companies total headquartered in New York as of June 2026. Those directories overlap imperfectly (different methodologies, different thresholds), but the signal is consistent: New York hosts a three-digit population of AI startups, and a meaningful slice of them build generative media tools.

Hedra, a New York–based generative-media company, builds around its own proprietary video model, Character-3, which it calls the first multimodal foundation model in production. The company has 14 employees and six open roles, three of them in engineering. That headcount is small, but Hedra's hiring pattern, ML engineers, AI/ML specialists, platform builders, mirrors what Mirage is posting. The two companies draw from the same shallow pool of video-model engineers in the five boroughs.

The broader New York generative-AI ecosystem feeds that pool from adjacent verticals. AlphaSense trains large language models on financial corpora. Kensho Technologies, an S&P Global subsidiary, employs 175 people across two New York offices and lists 22 open roles, 10 in engineering and 8 in AI/ML. Hebbia, backed by Andreessen Horowitz and Peter Thiel, has 28 open positions. These companies do not build video models, but they recruit the same research engineers, the people who understand distributed training, inference optimization, and evaluation pipelines, that generative-video startups need.

The talent competition is not theoretical. Mirage lists open roles in Union Square, three of them directly in generative video and agentic systems, at the top of the market range. Runway, Mirage's most direct competitor in generative video, has four open roles, all remote. The geographic split is telling: Runway hires from anywhere; Mirage hires from New York. That constraint is a bet that the city's density of media companies, ad agencies, and enterprise marketing teams gives it a recruiting edge for engineers who want to work near production clients.

New York's structural advantage is client proximity, not compute. The city has no meaningful GPU cluster advantage over San Francisco or Austin. But it houses the headquarters of major advertising holding companies, top film and TV post-production houses, and the finance and healthcare enterprises that buy AI-video tools. For a company like Mirage, whose Captions product already serves millions of creators and is converting that base into enterprise workflows, being in the same time zone as the buyers shortens the sales cycle and the feedback loop.

The risk is that San Francisco still wins the deepest technical talent. New York AI startups are growing fast on search volume, but the largest model-training labs (OpenAI, Anthropic, the teams building the next generation of video foundation models) remain concentrated in the Bay Area. New York's counter is that it captures a different segment: engineers who want to build applied generative-media products for enterprise buyers rather than foundational models for their own sake. Mirage's hiring, those same two ML roles, is aimed squarely at that segment.

The corridor is real, but it is narrow. New York has enough generative-media startups to create a local labor market, enough enterprise clients to justify the location, and enough university output from Columbia, NYU, and Cornell Tech to replenish the pipeline. What it does not have is depth. When Mirage needs a fifth video-model engineer, the local candidate pool is thin. That is the trade the company is making with its Union Square location: proximity to buyers over proximity to the largest possible talent supply.

What the CVF Structure Means for AI-Video Workforce Scaling

General Catalyst's Customer Value Fund doesn't behave like a traditional VC round, and that distinction reshapes how Mirage can hire. The CVF structure, non-dilutive capital repaid through a fixed, capped share of revenue generated from the funded spend, means Mirage's $75M is tied to a specific engine: sales and marketing. General Catalyst gets paid back only when that spend produces revenue. If it doesn't, GC absorbs the loss. Mirage doesn't owe a fixed repayment regardless of performance.

That has direct workforce implications. Because the CVF is designed to fund customer acquisition rather than product risk, Mirage's engineering hires sit on a different capital footing than its go-to-market hires. The ML engineers and generative-video researchers the company recruits are not funded by the CVF. They're funded by the equity and revenue Mirage preserves because it didn't dilute to fuel sales. The CVF model explicitly aims to let companies redirect equity capital toward "unstructured risk" like product and engineering.

This is the opposite of how most generative-video competitors operate. Runway raised traditional venture rounds where every dollar of sales-and-marketing spend came from equity that diluted the cap table. That model forces a tradeoff: fund growth or preserve ownership. The CVF splits the difference, which means Mirage can scale its sales organization aggressively while keeping engineering hires on a stable equity-backed budget.

The structure also affects go-to-market speed. Grammarly used its $1B CVF commitment, the largest in the fund's history, specifically for sales and marketing, freeing existing capital for the Coda acquisition. Musely took $360 million from the same fund to fuel customer acquisition it had been self-funding for a decade. In both cases, the CVF let a profitable company accelerate hiring in revenue-generating functions without touching its equity runway. Mirage's $75M signals the same logic applied to generative video: the company has enough product-market fit to predict a return on sales spend, and GC is underwriting that bet.

The risk is concentration. CVF deals require predictable revenue streams and strong unit economics, and GC's own materials describe the fund as targeting companies where customer acquisition cost and lifetime value are measurable enough to underwrite. If Mirage's enterprise video pipeline stalls, the repayment clock doesn't stop, even though GC bears the downside. For workforce planners, that means the sales and marketing roles the CVF funds carry a different kind of pressure than a traditional equity-funded ramp: they need to convert, and convert on a timeline that services the revenue share.

For engineers watching the generative-video job market, the takeaway is concrete. Mirage's current open roles skew heavily toward ML and agentic systems, the product-risk side of the business that equity, not the CVF, underwrites. That's where the company is placing its long-term bets. The CVF is fueling the engine that pays for those bets. If you're weighing offers across the generative-video stack, that capital-structure split matters: it tells you which hires the company expects to generate revenue next quarter and which ones it's building around for the next three years.

Working in AI? Zero G Talent tracks the openings: browse AI jobs, openings at OpenAI, Runway and Mirage, and the people building the field.