Captions AI / Mirage

1. Company Overview

2. Captions → Mirage Transition

On 4 September 2025, Gaurav Misra announced that the company was changing its name from Captions to Mirage. The announcement explicitly said the Captions app would remain central, while Mirage would represent a broader vision around frontier AI research and models. Current Mirage materials now make the hierarchy explicit: Mirage is the company; Captions is a flagship product.

3. Founders & Leadership

Gaurav Misra — Co-Founder & CEO

Mirage's official profile says Misra has a background in engineering, design and machine learning. Before Mirage, he led Design Engineering at Snap, where the profile says he oversaw social products including Spotlight and Snap Map. Mirage says Misra and Dwight Churchill met while working at Localytics and saw the gap between the rapid rise of short-form video and the difficulty of making it.

[VERIFIED]

Source: Mirage founder profile.

4. Founding Story

Mirage's official founder biographies say Misra and Churchill met at Localytics and noticed the rise of TikTok alongside a difficulty that remained unresolved: short-form video was becoming a dominant format, but making it was still hard. Misra's prior design-engineering work at Snap exposed him to the evolution of social video. The company launched in 2021 around the idea of reducing friction in short-form/talking-video creation.

Independent reporting from Forbes describes a 2021 prototype for a “talking-video” social network, followed by experimentation and a pivot toward the Captions product. By 2023, Captions reported millions of creators and a rapidly expanding creative toolkit; by 2024, the company had shifted decisively toward generative video research; by 2025, it adopted Mirage as the company identity and began positioning itself as an AI-video research/model company as well as the maker of Captions.

5. Company Timeline

6. Mission & Strategic Evolution

Analyst interpretation: the company is moving “up the stack” from a consumer editing interface toward proprietary models plus an editable creative runtime. That is an interpretation of the documented product sequence, not a stated financial outcome.

7. Current Captions Product

8. Current Mirage Platform

9. Mirage Foundation Models

UGC Video Foundation Model — March 2025

Mirage announced a proprietary foundation model for UGC-style video. The company said it generates people, objects and background environments that do not exist, and can be directed by optional text plus audio, video or script. It positioned the model as a step beyond lip-sync because it generates facial and upper-body motion rather than simply changing mouth movement on pre-existing footage.

Casting Voices — June 2025

The research report describes an audio-to-video foundation model for avatar video. It takes images, text and audio and aims to synchronize emotional nuance, lip movement and gesture so that the generated person appears to be delivering the supplied performance.

Finding Voices — December 2025

Mirage's audio research treats voice performance as a generative input rather than a neutral speech layer. The company says its Mirage audio model generates expressive vocal performances and is intended to pair naturally with generated video.

10. Mirage Avatar X

Avatar X was announced in July 2026 and is the current avatar-generation model publicly tied to Captions' avatar and AI Twin workflows. Mirage's own public materials claim:

• 10 seconds of input video for an AI Twin.
• Identity preservation and richer micro-expressions.
• Horizontal and vertical generation.
• Consistent quality across long-form/continuous generations.
• Audio, video and movement generated together for the AI Twin workflow.
• 200+ ready-made avatars in the Captions library, with prompt-generated custom avatars.

[COMPANY-REPORTED] Terms such as “industry-leading” and “new standard” are marketing claims and are not reproduced here as independent rankings.

11. AI Twin vs AI Avatar vs Custom Avatar

12. Mirage Audio & Voice

The company treats audio as a core generative modality. Current Captions documentation lists AI Voice Clone, AI Voiceover, AI Sound Effects and AI Denoise; the Mirage research program adds a proprietary expressive-audio direction.

The specific internal architecture details requested in the prompt—such as a verified use of diffusion, flow matching, latent-audio design, “Mochi Transformer,” or an autoregressive cascade—are not established by the reviewed public primary sources and are therefore not asserted as facts here.

13. AI Video Editing & Agentic Creation

Chat-based editing

In September 2025, Captions announced a conversational editor: users could type instructions such as adding a zoom, adding captions or applying a desired style, with the AI applying the edit.

Mirage Agent

Current Mirage materials describe Mirage Agent as a multimodal agent that can plan, create, edit and design video while directing proprietary and third-party models. This is materially broader than an “AI filter”: the intended workflow is natural-language direction across generation and editing.

Tesseract

September 2026 Tesseract updates say it supports editing, motion graphics, compositing and sound for agents, including Linux support and 4K/60fps export. The Tesseract site describes the project as editable, with keyframes, adjustment layers, compositions and audio layers exposed to agents.

14. Product Ecosystem

15. Users, Scale & Growth

Do not compare the creator and user figures as if they were identical metrics. The company changed terminology as the product broadened.

16. Funding & Investors

Total: Mirage reported total funding of more than $175M after the March 2026 financing. Because the earlier seed and Series A amounts were not publicly specified in the company's later announcements, the exact cumulative cash raised cannot be reconstructed from public primary disclosures alone.

Notable investors / backers documented across the financing history

Sequoia Capital, Andreessen Horowitz, Kleiner Perkins, Index Ventures, Adobe Ventures, HubSpot Ventures, SV Angel and General Catalyst's Customer Value Fund are documented investors. Earlier individual/angel backers publicly named by the company include Kevin Systrom, Mike Krieger, Lenny Rachitsky, Julie Zhuo, Antoine Martin, Nico Wittenborn, Jacob Andreou, Ben Rubin and Jaren Glover.

17. Funding Interpretation

18. Business Model & Pricing

Current Captions pricing is a freemium consumer/creator model plus higher-usage plans and custom enterprise packages. Pricing is dynamic and should be checked before publication.

Pricing checked against the current Captions pricing page on 30 September 2026. The page notes that displayed prices/features are for iOS plans.

19. Enterprise

Captions' enterprise positioning focuses on high-volume video production, centralized brand controls, distributed teams, approval workflows, custom plans, support and security. The enterprise page says most teams can be up and running within days and emphasizes brand templates and controlled standards.

20. Customers & Case Studies

Public customer evidence is not as extensive as the consumer-user figures. The strongest directly documented enterprise/customer examples reviewed are:

HubSpot deep dive

Problem: HubSpot was expanding paid campaigns and found strong UGC performance, but sourcing creators/participants and completing edits took too long; quality was variable and the team was stretched by creative fatigue.

Solution: Captions was positioned as a way to produce creator-style, on-brand UGC content with flexible customization and faster iteration.

Evidence: The official case study confirms the problem and solution, but the reviewed page excerpt did not provide a sufficiently precise numerical ROI metric to independently calculate a return percentage. Therefore ROI is not publicly calculable from that source alone.

Read the official HubSpot case study →

21. AI UGC Strategy

UGC is strategically important because Mirage's foundation-model work was explicitly designed around UGC-style ads and talking content. The company's March 2025 research describes a move away from licensing/casting pre-recorded actors toward generating the performer, environment and performance. Its enterprise positioning then connected the technology to paid creative testing and scalable ad production.

22. Marketing & Distribution Strategy

• Consumer/product-led growth: Captions is available as an app/web product with a free tier and paid upgrades.
• Creator economy: The company targets creators across beauty, fashion, fitness, food, podcasts, education and other verticals.
• Business/enterprise: Marketing teams, agencies, small businesses and enterprise teams are now explicit segments.
• AI-generated demonstrations: Mirage increasingly uses its own models and products as demonstrations of what the technology can produce.
• Asia expansion: The March 2026 funding announcement explicitly says Mirage planned aggressive expansion into Asia and characterized demand there as strong; country-specific growth figures were not publicly disclosed in that announcement.

23. Research & Technical Program

24. Technology Architecture — What Can Actually Be Verified

25. API / Developer Platform

Mirage currently exposes APIs for video generation and captioning. Its API legal terms describe Mirage API access and explicitly prohibit using the service or outputs to train, fine-tune, evaluate for training, or otherwise improve a competing AI/ML model. The current product ecosystem therefore has three layers: consumer app, agent/runtime tooling and developer APIs.

26. Security, Privacy & Responsible AI

SOC 2

Mirage announced on 9 March 2026 that it completed a SOC 2 audit, with independent auditors reviewing its security, availability and data-protection controls.

Infrastructure

Mirage's safety page says its systems are hosted in isolated, U.S.-based Google Cloud data centers, with encryption, role-based permissions, audit trails, MFA, continuous scanning, third-party penetration tests and 24×7 monitoring.

Enterprise training-data policy

Mirage's enterprise page states that enterprise data is never used to train Mirage's models and that enterprise data remains private and secure. Current pricing also lists “training data exclusion” as an enterprise feature.

Biometric information

The November 2025 privacy policy says Mirage may generate biometric information from face or voice for certain features. It states that biometric identifiers/information are deleted at the earlier of the initial purpose being satisfied or three years after the user's last interaction, subject to legal retention requirements.

Consent and likeness

The current Terms require users to have written consent from identifiable people whose name, likeness or voice appears in Input. The policy also restricts impersonation and deceptive use.

27. Legal, Copyright & Content Governance

• Mirage's Terms state that users own rights in their Input and Output as between the parties, subject to Mirage/licensor rights in included content.
• Users grant Mirage a license to host, store, reproduce, modify, adapt, publish, translate, distribute and use Input/Output to provide and improve services and develop new products.
• The Terms specifically state that Mirage will not commercialize a user's voice on a standalone basis without written permission when the Input contains a voice or likeness.
• The AUP prohibits deceptive provenance claims, disinformation campaigns, impersonation, unlawful/infringing uses and other harmful conduct.
• The current web terms require disclosure that AI was used to generate Captions on-web output when appropriate and require advertising disclosures where law requires them.

Litigation: No specific major lawsuit or court dispute is documented in the reviewed primary sources. This report is not a comprehensive legal-docket search, so absence from this report must not be interpreted as proof that no dispute exists.

28. Current Users & Market Segments

29. Competitive Landscape

The company operates across several categories rather than one narrow market. Current documented alternatives include CapCut, Adobe's video ecosystem, Descript, OpusClip, Runway, HeyGen, Synthesia, Canva, VEED, InVideo and ElevenLabs. Their overlap varies: some are primarily editors, some generative-video platforms, some avatar/video communication systems, and some audio/model providers.

No competitor is ranked here. The table is a factual category comparison, not a recommendation.

30. Captions vs CapCut / HeyGen / Descript — Factual Framing

A full current pricing/capability matrix against every competitor would require separate primary-source verification for each competitor and is intentionally not fabricated here.

31. Global Expansion

The company has consistently used localization as a product feature. The 2023 Lipdub launch supported 28 languages. Current AI Creator documentation supports 30+ languages, and current Captions pricing lists captions in 100+ languages. The March 2026 financing announcement says Mirage plans to expand aggressively into Asia. Specific country-by-country user counts are not publicly disclosed in the reviewed primary sources.

32. Acquisition

In November 2024, Axios reported that Captions acquired AlpacaML, a generative-video company. The acquisition was framed as strengthening the company's AI expertise and technology base. The public report does not establish a detailed post-acquisition org chart or full technology integration map.

33. Major Achievements

• Reached 10M+ global creators and 3M+ monthly videos by July 2024, according to the company.
• Raised $60M Series C at a $500M valuation in July 2024.
• Committed to $100M of generative-video research investment in New York.
• Published proprietary foundation-model research for UGC video and avatar video.
• Rebranded the company from Captions to Mirage in September 2025 while retaining Captions as a flagship app.
• Reached a company-reported 20M+ global users and 250M+ videos by March 2026.
• Raised $75M growth financing from General Catalyst's Customer Value Fund in March 2026.
• Completed a SOC 2 audit in March 2026.
• Launched Avatar X in July 2026.
• Expanded Tesseract with Linux, 4K/60fps export and additional agent integrations in September 2026.

34. Challenges & Risks

35. Responsible AI

Mirage says it uses automated and human moderation, stress-tests systems against disallowed content and layers safeguards at model, application and platform levels. Its AUP prohibits deceptive provenance, impersonation and disinformation uses. The Terms require consent for identifiable people's voice/likeness in Input. These are documented safeguards; they should not be interpreted as proof that misuse is impossible.

36. Important Company Numbers

37. SWOT — Evidence-Based

38. Strategic Insights

1. Why do companies buy/use Captions? To reduce the friction of making frequent video content: editing, captions, UGC-style creative, avatars, localization and content variations.
2. What department appears first? Public evidence is strongest around creators and marketing teams; enterprise expansion evidence is still thinner than consumer-scale evidence.
3. Why does the company own foundation models? The 2024–2026 research program suggests the company believes model-level control can improve video performance and enable product capabilities that generic editors cannot provide.
4. Is localization important? Yes. Dubbing and multilingual generation have been part of the product evolution since at least 2023.
5. What is the biggest strategic change? The move from “AI features inside an editor” to “AI as the director and editor of a complete creative production.”
6. What is Tesseract's significance? It potentially turns creative editing primitives into tools that AI agents can operate directly, reducing dependence on GUI navigation.
7. What remains unsolved? Public evidence does not establish a complete autonomous creative pipeline that eliminates human review; the company's own emphasis on direction, editing and refinement indicates that human judgment remains important.

39. Current 2026 Status

40. Final Assessment

41. Key Facts

• Mirage says it is the company formerly known as Captions.
• Captions is a flagship Mirage product, not a discontinued legacy brand.
• Gaurav Misra and Dwight Churchill are the co-founders.
• Gaurav Misra is CEO; Dwight Churchill is COO.
• The company is based in New York City.
• The company raised $60M Series C in 2024 at a reported $500M valuation.
• Mirage announced $75M of growth financing from General Catalyst in March 2026 and reported total funding above $175M.
• Mirage reported 20M+ global users and 250M+ videos by March 2026.
• Mirage's current research program spans UGC video, avatar video and expressive audio.
• Avatar X is the current proprietary avatar model highlighted in Captions.
• Tesseract is the current agent-oriented creative runtime.
• Mirage completed a SOC 2 audit in March 2026.

42. Business Lessons

1. Distribution can finance research: The company used a large creator product to build a user base before making proprietary model research a central strategic priority.
2. Product and model layers can reinforce each other: Captions provides distribution and feedback; Mirage models provide differentiated generation/editing capabilities.
3. AI agents need native creative primitives: Tesseract represents an attempt to expose editing, motion and sound directly to agents rather than forcing agents to operate a human GUI.
4. Identity is a product surface: AI Twins make likeness, voice and mannerisms reusable production assets, but that increases the importance of consent and safety controls.
5. Short-form video is a useful model-training target: Mirage's 2025 research explicitly focuses on pacing, framing and attention dynamics for short-form content.

43. Source Audit

44. Evidence & Uncertainty Guide