Captions, the popular AI video app for creators, is changing its name to Mirage and repositioning itself as an AI research lab focused on short-form video. The company, which has raised more than $100 million and was last valued around $500 million, says the new identity reflects a shift from building tools to developing foundational multimodal models optimized for TikTok, Reels, and Shorts.
The rebrand consolidates its consumer app (formerly Captions) and its business-facing product, Mirage Studio, under a single umbrella. The bet: purpose-built models for vertical, speech-led video can outperform general-purpose systems and give brands a faster, cheaper way to generate ads without live shoots.
From creator app to research lab
Mirage is orienting its R&D around multimodal generation—models that jointly reason over text, audio, and video—to better map speech to facial motion, body language, and scene dynamics. Unlike large, generic video models developed by the biggest AI labs, Mirage is narrowly focused on the constraints that define short-form: portrait framing, punchy pacing, branded callouts, and latency low enough to iterate quickly.
That specialization matters. Short-form ads live or die on micro-signals like lip articulation, eye gaze, and gesture timing. A model tuned for natural prosody and 9:16 composition can deliver higher completion rates and lower cost per view than legacy pipelines that stitch together stock footage or rely on brittle lip-sync. Mirage’s thesis is that verticalized models can capture these nuances while keeping inference costs within the realities of performance marketing.
What Mirage Studio actually does
Mirage Studio targets brand and agency workflows. Upload a script or audio, select a style, and the system generates an ad-length clip—complete with an AI actor, a synthetic background tailored to the brief, and on-brand motion graphics. Users can also create an avatar from their own likeness, with explicit consent, for scalable spokesperson content.
Pricing sits at $399 per month for 8,000 credits under a business plan, with a promotional 50% discount for the first month for new customers. The pitch is simple: reduce time-to-launch and production overhead while maintaining control over voice, look, and messaging.
Mirage differentiates by claiming end-to-end generation rather than assembling clips from stock libraries or using shallow lip-sync over prerecorded video. The company emphasizes lifelike speech dynamics and facial expressions as core performance features, not add-ons.
The market context
Short-form video is where attention has pooled. Industry analysts at data.ai have reported that users in many markets spend more time in TikTok than in any other social app, while Insider Intelligence has projected U.S. digital video ad spending to approach the $100 billion mark as brands chase that engagement. For performance marketers, the appeal of synthetic production is less about novelty and more about speed: crank out 50 variants, measure lift by cohort, double down on the winners.
The field is crowded. Avatar and presenter platforms like Synthesia, D-ID, Hour One, and HeyGen have established beachheads in corporate training and product explainers. Generative video labs such as Runway and Pika focus on cinematic and creator-led workflows. Mirage is carving out the short-form ads segment with a research-first posture and a full-stack approach that spans modeling, studio tooling, and brand safety controls.
Risk, provenance, and policy
Any company advancing realistic synthetic video faces the deepfake dilemma. Academic work from institutions including MIT Media Lab and UC Berkeley has shown that unaided humans struggle to reliably spot manipulated video, often performing barely above chance. That’s a problem in an era when deceptive clips can travel far before fact-checkers catch up.
Mirage says it restricts impersonation, requires consent for likeness use, and applies moderation to curb misuse. The company also argues that design alone won’t solve misinformation, calling for stronger media literacy so audiences treat video claims as skeptically as headlines. In parallel, industry efforts are gaining traction: the C2PA standard for content credentials—backed by organizations like Adobe, Microsoft, and the BBC—aims to embed provenance signals in media files, while major platforms have begun labeling AI-generated content. Pressure will mount on vendors like Mirage to implement robust provenance by default and to integrate with platform-level authenticity checks.
What to watch next
Three questions will determine whether Mirage’s pivot pays off. First, can its specialized models demonstrably beat general-purpose systems on short-form ad metrics such as watch-through rate and cost per acquisition? Second, will the economics of inference—GPU costs, throughput, and latency—support wide-scale deployment for brands running constant creative tests? Third, can Mirage maintain strict safeguards without throttling the flexibility advertisers want?
If the company ships state-of-the-art research that translates into measurable performance lift, integrates provenance standards, and keeps production costs predictable, it will be well positioned in a crowded but fast-growing segment. The rebrand to Mirage signals ambition beyond creator tooling; now the lab’s outputs will have to prove it.