Captions Mirage

Captions Mirage is the research and foundation-model layer behind Captions, built to generate and edit short-form audiovisual content from an idea through a finished video.

Verified capabilities and evidence

Mirage says its models treat text, audio, and video together, learning phrasing, vocal cadence, pose, micro-expressions, framing, emotion, and scene structure. Its research page describes in-house training and inference plus a curated, licensed audio-video dataset designed around rights, safety, and brand control. The Captions pricing page names AI Creator, AI Ads, AI Skits, AI Twin, AI Edit, Clips, AI Translate, AI Lipdub, AI Music, AI Sounds, AI Voiceover, AI Image, and AI Video. Captions support more than 100 languages, and paid plans include over 100 caption templates.

Workflow and differentiation

A creator can generate footage with an AI actor or digital twin, edit through chat, add custom B-roll, images, music, and sound effects, then translate or lip-sync the result for another market. Long recordings can become batches of post-ready clips. Conventional operations such as trim, zoom, keyframes, transitions, resizing, transcripts, teleprompter use, voice cleanup, and captions sit beside generation. That integrated workflow is the main difference from combining an avatar generator, caption service, timeline editor, and stock library. Brand-sensitive output still needs a human final edit.

Pricing and total cost

The official page says displayed prices are USD iOS plans. Free has traditional editing but no AI credits. Max is $24.99 per month for 500 credits. Scale is $69.99 for 1,400 credits; Scale 2x is $139.99 for 2,800; Scale 4x is $279.99 for 5,600. Enterprise is custom and includes bulk credit discounts, custom seats, training-data exclusion, onboarding, priority training, and early beta access. Credit cost per output is not fixed on the page, so teams must measure usable seconds per credit for their own format.

Pros, cons, and evidence gaps

Captions covers unusually much of the short-video pipeline and gives non-editors concrete shortcuts. A unified multimodal model may synchronize script, delivery, expression, and framing better than loosely connected tools. Limits matter: generated people can look or move unnaturally, translations and lip sync can misstate names or claims, and 500 credits is not a guaranteed number of finished videos. Listed prices are explicitly iOS prices, so web, annual, tax, and regional billing may differ. Custom actors and voices also create consent and likeness risks.

Practical evaluation plan

Create ten real briefs across a talking-head explainer, product ad, and translated social clips. Measure usable seconds per credit, regeneration rate, lip-sync quality, caption corrections, rendering time, and final editor minutes. Obtain consent for every custom actor and voice. Compare avatar output with Synthesia, broader design work with Canva, and voice options with ElevenLabs. Teams creating repeatable publishing operations should also read the social media automation guide and document who approves factual claims before release.