Captions, the AI video app for creators, is rebranding as Mirage and repositioning itself as an AI research lab that’s centered around short-form video. The company, which has raised over $100 million in funding and was last valued at around $500 million, says the rebranding aims to better describe the company’s pivot away from building tools to power functions in social video apps, like TikTok, Reels, and Shorts, to instead developing foundational multimodal models that can be applied across platforms.
The move unites its consumer app (previously Captions) and the business-focused Mirage Studio product under one roof. The bet: Model purpose-built for vertical, speech-led video can outperform general-purpose systems and offer a faster, cheaper way for brands to create ads without live shoots.
From creator app to research lab
Mirage is centering its R&D on multimodal generation—models that reason jointly over text, audio, and video—so that it can map speech more accurately to facial motion, body language, and scene dynamics. Unlike large, generic video models from the largest AI labs, Mirage is specifically designed with the constraints that shape short-form: portrait framing, punchy pacing, branded callouts and latency low enough to iterate rapidly.
That specialization matters. Micro-signals like lip articulation, eye gaze, and the timing of gestures determine the fate of short-form ads. The natural prosody and 9:16 composition bespoke model can achieve higher completion rates and lower cost per view versus previous pipelines using stitched stock footage or brittle lip-sync. Mirage’s premise is that verticalized models can capture these idiosyncrasies all while ensuring that inference is affordable in the context of performance marketing.
What Mirage Studio Actually Does
Company summary Mirage Studio brand and agency solution is the choice for your workflows. You upload a script or audio, pick a style, and the system automatically spits out an ad-length clip — complete with an AI actor, a synthetic background designed to fit the brief and on-brand motion graphics. Obviously with consent, users can also create an avatar of themselves for scalable spokesperson content.
Pricing is $399 per month for 8000 credits on a business plan, although new customers get a 50% discount for the first month. It’s an easy pitch: cut time-to-launch and production overhead, and retain control of the voice, look, and message.
Mirage stands out by boasting end-to-end generation rather than stitching together clips from stock libraries or doing bare-minimum lip syncing over already recorded video. The firm highlights naturalistic speech dynamics and facial expressions as key performance abilities rather than add-ons.
The market context
That’s where attention has cohered. Industry analysts at data. AI have reported that in many markets, users spend more time in TikTok than in any other social app, and Insider Intelligence projected that U.S. digital video ad spending will reach nearly $100 billion as brands chase that engagement. For performance marketers, what’s intriguing about synthetic production isn’t that it’s new, but that it’s fast: You can just crank out 50 variations, measure lift by cohort, and 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. Operations by generative video labs like Runway and Pika center on a cinematic and creator-driven approach. With a research-first stance and a full-stack model that covers modeling, studio tooling and brand-safety controls, Mirage is staking a claim in the short-form ads space.
Risk, provenance, and policy
Any company pushing the state of the art in realistic synthetic video runs up against the deepfake dilemma. Academic studies from institutions like MIT Media Lab and UC Berkeley have also demonstrated that unassisted humans are bad at spotting fake video, with our accuracy only slightly better than random. That’s a challenge at a time when false clips like that can go viral long before fact checkers can catch up.
Mirage says it prohibits impersonation, mandates consent for using one’s likeness, and enforces moderation to deter abuse. The company also contends that design by itself will not fix the misinformation problem and that what is needed is more robust media literacy, so that when people see a claim in a video, they approach it with the same skepticism as when they read a headline. Meanwhile, industry initiatives are gathering momentum: the C2PA standard for content credentials, supported by companies including Adobe, Microsoft and the BBC, seeks to bake provenance signals into media files, and major platforms have started flagging AI-generated content. Vendors like Mirage will increasingly be called on to implement strong defaults for provenance and integrate with platform-level authenticity checks.
What to watch next
Three questions will determine whether Mirage will succeed in its pivot. One, can its specialized models convincingly outperform general purpose systems on short form ad metrics like watch through rate and cost per acquisition? Second, will the economics of inference – GPU costs, throughput, and latency – support broad-based deployment for brands with constant creative testing? Third, can Mirage adhere to rigorous protections that don’t throttle the freedom advertisers are after?
If the company can deliver new research that can be attributed to a solid performance lift, integrate provenance standards and keep costs in production predictable, it will be in a good place among an increasingly crowded but rapidly growing segment. The rebrand to Mirage signals an ambition beyond creator tooling; to what degree the lab’s output will demonstrate this ambition will now have to be seen.