Captions rebrands to Mirage to build short-form video AI models

Short-form video is eating the internet, and AI avatars are lining up to replace creators in front of the lens.

Captions, the video editing platform valued at $500 million after raising more than $100 million, just announced on Thursday that it is rebranding to Mirage. But this is more than a name change: the company is expanding beyond surface-level creator apps into a dedicated AI research lab building multimodal foundational models for TikTok, Reels, and YouTube Shorts.

As CEO Gaurav Misra put it, the team believes the real race for AI video hasn't even begun yet.

Beyond lip-syncing and stock footage

The rebrand brings everything under one roof, uniting the original creator app with Mirage Studio, an enterprise platform launched in June designed specifically for ad production.

The premise of Mirage Studio is simple: brands can churn out video ads without hiring talent or burning production budgets. You upload an audio file, and the AI builds the entire video from scratch, generating custom backgrounds and avatars. You can even generate a digital clone just by submitting a selfie.

Unlike traditional tools from competitors like Synthesia, D-ID, or Hour One, Mirage claims it doesn't rely on existing stock footage, voice cloning, or basic lip-syncing tricks. Instead, it says its models generate natural gestures, facial expressions, and speech natively.

Access doesn't come cheap. Mirage Studio lives inside the company's $399-per-month business plan, which includes 8,000 credits, though new accounts get 50% off their first month.

The backlash against synthetic talent

Cutting human creators out of the loop always ruffles feathers. The advertising industry is already facing friction over virtual humans, highlighted by the public blowback after Guess featured an AI-generated model in Vogue's July print issue.

Then there is the unavoidable deepfake dilemma. As visual synthetic artifacts disappear, identifying deceptive video gets nearly impossible for casual viewers.

Mirage acknowledged these risks in a recent post, noting that it enforces moderation checks like preventing impersonation and requiring consent before using someone's likeness. However, the company admitted that engineering guardrails alone won't solve the problem, arguing that the public needs to develop a new level of media literacy to scrutinize video just like text.

Why it matters

Workflow tools come and go, but model ownership gives you real leverage. By stepping away from the pure editing wrapper model and investing in proprietary foundation models, Mirage is attempting to leapfrog legacy avatar players who mostly rely on clever stitching and lip-syncing overlays.

For performance marketers and bootstrapped brands, the value proposition is obvious. Testing dozens of variations on TikTok without booking studio time saves thousands of dollars, making a $399 monthly subscription an easy line item to justify.

Focusing exclusively on short-form vertical video is a shrewd move. TikTok feeds move so fast that viewers are far more forgiving of micro-glitches than they would be while watching a feature film.

Still, telling users that the real solution to deepfakes is better public media literacy feels like passing the buck. As synthetic video flood feeds, regulators and platform distribution channels are likely to hold companies like Mirage accountable for the content they enable, whether the avatars look human or not.