Artificial Intelligence Files · Research Edition

Captions AI
Real-World Case Studies

An evidence-led database of documented Captions users, customer stories, creator workflows, AI-generated advertising, editing, localization and measurable outcomes — with company claims separated from independently verified evidence.

Research cut-off: 30 Sep 20266 detailed documented casesPrimary-source firstNo invented metrics

Generic product examples, logo walls, demos and unnamed use cases are not promoted to confirmed customer case studies.

Gaurav Misra — supplied Artificial Intelligence Files image
Captions logo — supplied Artificial Intelligence Files asset
20M+global Captions users reported by Mirage in March 2026.
01 · Identity & evidence frame

Captions is the app. Mirage is the company.

[VERIFIED]

Current corporate identity

Mirage’s current corporate site says the company was formerly known as Captions and now builds the Captions app, Mirage foundation models and the Mirage API. The September 2025 announcement says the company changed its name from Captions to Mirage while keeping the Captions app central to its mission.

A Mirage model demonstration is therefore not automatically a Captions customer case study.

[CURRENT]

What the product does

Current documentation describes Captions as an AI video creation/editing product with AI Edit, captions, dubbing, AI avatars/twins, generated media and Mirage-powered generation.

Mirage About · Rebrand announcement · Captions Help

02 · Executive summary

What the documented evidence shows

6

strongly documented cases form the core database in this edition.

4

quantified performance stories contain explicit speed, cost or ad metrics.

3

2026 case stories are in the core set: Peter, Bonsa and the current Captions/Mirage context.

0

independent ROI benchmarks can be calculated from fully disclosed public investment data.

Evidence boundary: Mirage’s March 2026 financing announcement names HubSpot, CoreWeave and King as marketing-team users. Only HubSpot has a detailed public case study in the reviewed evidence, so CoreWeave and King remain customer mentions rather than detailed case studies.
03 · Master case database

Documented customer and creator stories

Customer / creatorIndustryTypePrimary useFeaturesDocumented resultEvidence
HubSpotSaaS / MarketingEnterpriseUGC performance campaignsAI characters; variations; multilingual generationMore output; lower reported cost/time; faster testingB
PeterReal estateSmall businessPersonal-brand video / lead generationTeleprompter; AI Edit; Prism Pro; generated images>2× views; under 1h finished video; listing opportunityB
BonsaCreator economyIndividual creatorSocial video productionAI Edit; automated finishing15-min finish; nearly $100K generated through contentB
Anonymous startupPerformance marketingStartupUGC ads / creative testingAI proof-of-concepts; hooks; CTAs; localization1-day launch; +83% CTR; +33% CVR; +41% hook rateD
PlugAd tech / E-commerce enablementStartupPaid-social creative at scaleAI Creators; localized AGC1 month→1 day; 5× lower CPL; 50% lower production costB
Elise MichealsCoaching / CreatorIndividual creatorCaptioning and batch publishingAuto transcription; captions; silence trimming~5 min captioning/video; engagement increase reportedB

Evidence scale: A = independently verified; B = strong primary evidence; C = customer testimonial; D = company-reported; E = third-party/creator report; F = promotional/demo.

04 · Detailed cases

Six evidence-rich case studies

CASE 01 · Enterprise customer case study
HubSpot — performance UGC at scale
SaaS / Marketing28 Oct 2025B · Strong primary evidence
DOCUMENTED
100svariations described
Fastercreative testing
MoreUGC-style output
Multimarket variants

Original problem

HubSpot’s paid-media team found strong UGC effective but difficult to source, shoot and edit at the required speed. Variable quality and creative fatigue constrained testing.

Captions solution

The team used Captions/Mirage to generate UGC-style performance assets, iterate characters and hooks, and create multilingual variants across markets.

Features used

Prompt-driven video generation; AI-generated characters; rapid variations; hook/outfit iteration; multilingual generation.

Business result

Higher output, lower reported production cost and turnaround time, and faster creative testing. No numeric ROI or conversion uplift is published.

BEFORE CAPTIONS

Participant sourcing → shooting → editing → handoffs → slower testing. Precise old cost/cycle time is not disclosed.

AFTER CAPTIONS

Generate and iterate directly in Captions; make hundreds of variations faster than a traditional single-video workflow; localize across markets.

“It’s hands-down the best text-to-video that I have seen for short-form video...”
— Kipp Bodnar, CMO, HubSpot
Causality check: The case attributes workflow improvements to integrating Captions; no independent controlled experiment is supplied.

Source: Captions — How HubSpot built better campaigns with Mirage

CASE 02 · Individual customer story
Peter — real-estate video that became a sales touchpoint
Real estate / Small business21 May 2026B · Strong primary evidence
DOCUMENTED
≤4hbefore / 2-min video
<1hfinished video
>2×views reported
1listing opportunity described

Original problem

A two-minute video could take Peter up to four hours to film and edit. He needed a professional presence but lacked time and editing skill.

Captions solution

He started with the teleprompter, then used automatic editing, generated images and Prism Pro to create consistent, polished real-estate videos.

Features used

Teleprompter; AI editing; generated images; Prism Pro; transitions and tighter edits.

Business result

Views more than doubled. A neighbor discovered Peter through a local community page and later contacted him for a listing presentation; Peter explicitly attributed the opportunity to the videos.

BEFORE CAPTIONS

Up to four hours for a two-minute video; manual scripting, filming and editing.

AFTER CAPTIONS

A polished finished video in under one hour, with consistent style and less manual editing.

“It looks like I’m a big polished team,”
— Peter, real-estate agent
Causality check: The view increase is reported by the case. The listing opportunity is customer attribution, not an independently isolated sales experiment.

Source: Captions — Peter case study

CASE 03 · Creator case study
Bonsa — dental creator turns editing time into publishing consistency
Creator economy30 Jun 2026B · Strong primary evidence
DOCUMENTED
15 minfinish time
~$100Kgenerated through content
Repeatablecontent format
Growingaudience, qualitative

Original problem

As Bonsa’s audience and quality expectations grew, transitions, text treatments and finishing details made every video labor-intensive.

Captions solution

He uses Captions after assembling footage and voiceover. Captions handles transitions, text treatments, auto-zooms and finishing; AI Edit helps shape difficult edits.

Features used

AI Edit; automated finishing; transitions; text treatments; auto zooms.

Business result

Once footage and voiceover are ready, he can finish a video in as little as 15 minutes. In recent months he generated nearly $100,000 through content.

BEFORE CAPTIONS

Manual editing of transitions, text and finishing details could take hours.

AFTER CAPTIONS

As little as 15 minutes to finish a video after footage and voiceover are ready.

“Captions does the final stretch for me.”
— Bonsa, creator / dental student
Causality check: The time reduction is tied to his workflow. The ~$100K figure is a broader creator-business result and is not presented as Captions-attributed revenue.

Source: Captions — Bonsa case study

CASE 04 · Anonymous customer case study
Fast-growing startup — performance creative iteration
Performance marketing2 Sep 2025D · Company-reported
ANONYMOUS
2+ weekssingle video before
1 daycampaign launch
+83%CTR
+33% / +41%CVR / hook rate

Original problem

The startup relied on UGC for performance marketing but took more than two weeks to make one video; creative fatigue arrived before new creative could launch.

Captions solution

Captions supplied an end-to-end UGC-style workflow: AI proof-of-concepts, iteration, new hooks/CTAs and faster localization across markets.

Features used

AI-generated proof-of-concepts; UGC-style generation; rapid hook/CTA iteration; localization.

Business result

Campaigns launched in one day; finished videos in less than a day; the latest creative was reported at +83% CTR, +33% conversion rate and +41% hook rate versus traditional UGC.

BEFORE CAPTIONS

2+ weeks per video; nearly 3 weeks for campaign launch.

AFTER CAPTIONS

One-day campaign launch; finished video in less than a day.

Causality check: The source attributes the comparison to the Captions workflow, but no independent experiment or named advertiser is supplied.

Source: Captions — startup case study

CASE 05 · Named customer case study
Plug — AI-generated ad creative for e-commerce clients
Ad tech / E-commerce enablement9 Dec 2024B · Strong primary evidence
DOCUMENTED
1 month→1 daydelivery turnaround
5× lowerCPL
50% lowerproduction cost
Localizedad variants

Original problem

Plug used UGC for client paid-social ads, but traditional UGC had long lead times, high costs, difficult localization and slow creative refresh cycles.

Captions solution

Plug used Captions AI Creators to generate localized, hyper-realistic ad content from prompts and re-generate hooks based on performance data.

Features used

AI Creators; prompt-based generation; localized ad content; hook iteration; avatar-generated content (AGC).

Business result

Plug reported one-day ad delivery, 5× lower cost per lead and 50% lower production costs after using Captions in campaigns.

BEFORE CAPTIONS

Ad delivery could take about one month; manual production and localization were expensive.

AFTER CAPTIONS

Reported one-day delivery, faster iteration and lower production costs.

Causality check: The case directly associates the campaign results with Captions AI Creators, but the underlying media-buying experiment is not independently audited in the source.

Source: Captions — Plug case study

CASE 06 · Historical creator customer story
Elise Micheals — captioning and batch publishing for a personal brand
Creator economy / Coaching22 Aug 2022B · Historical primary evidence
HISTORICAL
~5 mincaptioning / video
5videos / filming session
↑engagement reported
3major platforms cited

Original problem

Manual transcription and caption timing took too long. Elise often skipped captions despite viewers who were hearing-impaired or watching without sound.

Captions solution

She filmed batches, did basic edits in InShot, then used Captions to transcribe, remove silences/ums, time captions and prepare platform-ready drafts.

Features used

Automatic transcription; automatic captions; silence/pause trimming; caption styling; multi-platform export.

Business result

About five minutes of captioning per video; typically five videos per filming session; engagement increase reported by Elise.

BEFORE CAPTIONS

Manual transcription and timestamp alignment; captioning often skipped because it was too time-consuming.

AFTER CAPTIONS

About five minutes per video for the captioning process and easier batch publishing.

“My engagement has absolutely increased since I’ve started captioning my videos — and even more so since I switched to Captions.”
— Elise Micheals
Causality check: Engagement improvement is customer-reported; the case does not isolate Captions from other changes in content or audience.

Source: Captions — Elise Micheals case study

05 · Before → after

How the workflows changed

Traditional bottleneck

Human-heavy production

  • Source creators or participants.
  • Schedule shoots and collect footage.
  • Manually cut, caption and polish.
  • Repeat production for new markets.
  • Refresh creative after performance data arrives.
AI-assisted workflow

Generate → edit → iterate → localize

  • Prompt or upload source material.
  • Generate actors, video, B-roll or edits where appropriate.
  • Apply AI Edit or a style such as Prism Pro.
  • Change hooks, CTAs, characters or language variants quickly.
  • Keep human review for strategy, brand, accuracy and approval.

Before/after metrics database

CompanyMetricBeforeAfterImprovementEvidence
PeterTime for 2-min videoUp to 4hUnder 1hAt least ~75% shorter [CALCULATED]Source
BonsaFinish timeHoursAs little as 15 minExact percentage not calculableSource
Anonymous startupCampaign launchNearly 3 weeks1 dayEndpoints differ in unit; no percentage assertedSource
PlugAd delivery turnaround1 month1 day~96.7% shorter [CALCULATED]Source
EliseCaptioningManual / often skipped~5 min/videoPrevious numeric time not disclosedSource

Calculated values are arithmetic transformations of disclosed endpoints; they are not independent performance measurements.

06 · Business impact

Where measurable value appears

Production velocity

Peter, Bonsa, the anonymous startup and Plug all report substantially faster production or delivery.

Creative testing

HubSpot and the anonymous startup describe rapid iteration of hooks, characters and CTAs. The startup reports +83% CTR, +33% conversion rate and +41% hook rate for its latest creative.

Cost control

Plug reports 50% lower ad-production cost. HubSpot reports reduced production cost without publishing a percentage.

Localization

HubSpot and Plug connect Captions to multilingual/localized creative. Exact language counts are not disclosed in the reviewed cases.

Creator consistency

Bonsa and Elise show the value of repeatable publishing with less finishing work.

Lead / sales touchpoints

Peter’s story links professional video to a listing-presentation opportunity through customer attribution.

07 · ROI

What can and cannot be calculated

NOT PUBLICLY CALCULABLE

Independent ROI

Public case studies generally disclose outcomes but not the full Captions investment, baseline production economics and counterfactual. A defensible independent ROI percentage is therefore not publicly disclosed.

[CALCULATED]

Operational transformations

Where both endpoints are explicit, arithmetic transformations can be shown without calling them ROI. Plug’s one-month-to-one-day turnaround is approximately a 96.7% reduction in elapsed time if one month is treated as 30 days.

08 · Use-case analysis

What customers actually use Captions for

AI UGC & paid social

HubSpot, Plug and the anonymous startup are the strongest documented examples: creator-style ads, rapid variation and performance iteration.

AI editing

Peter and Bonsa demonstrate AI-assisted finishing from recorded footage; Elise demonstrates earlier automated transcription/captioning.

Personal-brand video

Peter and Elise show how a single operator can reduce the burden of polished social video.

AI generation

HubSpot and Plug use generated actors/UGC-style assets. Current Mirage documentation supports generated actors and video, but generic capability is kept separate from customer proof.

Localization

HubSpot and Plug explicitly connect Captions to multilingual/localized creative; no exact language count is asserted.

Lead generation

Peter’s story is the clearest public example of video supporting a real-world sales conversation.

09 · Feature → case mapping

Which capabilities have customer evidence?

FeatureCustomerUseResultStatus
AI EditBonsa; PeterAutomated finishing, transitions, text and pacing15-min finish for Bonsa; under-1h video for PeterPRIMARY
Prism ProPeterConsistent real-estate video styleUnder-1h finished video; views more than doubledCUSTOMER-REPORTED
AI Creators / AI actorsPlug; HubSpotUGC-style advertisingPlug: 1 month→1 day, 5× lower CPL, 50% lower cost; HubSpot: higher output / faster testingPRIMARY
AI-generated charactersHubSpotPerformance creative iterationDifferent characters, hooks and variantsPRIMARY
Multilingual generationHubSpotCross-market variantsLower reported cost/complexityPRIMARY
Auto captions / transcriptionEliseAccessible social video~5 min/video; engagement increase reportedHISTORICAL
Mirage Studio / Ad StudioProduct capabilityGenerated actors and adsCurrent capability; no separate named customer metric in reviewed evidencePRODUCT / NOT CASE
10 · Human + AI workflow

AI accelerates production; humans retain creative judgment

Human contribution

Strategy & judgment

  • Choose audience, message and creative goal.
  • Write/refine scripts, hooks and CTAs.
  • Decide brand tone, claims and visual direction.
  • Review output and correct errors.
  • Select winning creative and approve publication.
AI contribution

Production acceleration

  • Transcribe and caption footage.
  • Trim pauses and repetitive edits.
  • Apply AI Edit styles and visual treatments.
  • Generate actors, footage, images or variants.
  • Translate/localize where supported.
  • Iterate hooks and CTAs faster.
11 · Enterprise evidence

Large-company mentions need a higher evidence bar

OrganizationCurrent source saysClassification hereReason
HubSpotDetailed case study describes performance-marketing use of Captions/Mirage.CASE STUDY CUSTOMERDirect, detailed first-party case evidence.
CoreWeaveMirage’s March 2026 financing announcement names CoreWeave among marketing teams using Captions.CUSTOMER MENTIONNo detailed customer case located in the reviewed evidence.
KingMirage’s March 2026 financing announcement names King among marketing teams using Captions.CUSTOMER MENTIONNo detailed customer case located in the reviewed evidence.

A corporate mention is intentionally not upgraded to a detailed customer case without stronger evidence.

12 · Historical vs current

The product changed materially

2022

Caption-first creator workflow

Elise’s case centered on transcription, caption styling, silence removal and multi-platform publishing. Historical product evidence.

2024

AI Creators become an advertising workflow

Plug’s case documents AI-generated creator-style ad content, localization and performance-driven iteration.

2025

Mirage rebrand + generative UGC

The company changed its name to Mirage while retaining the Captions app. HubSpot documents generated characters, variations and multilingual performance creative.

2026

Current Captions + Mirage stack

Current documentation describes AI editing, generated actors/video, dubbing, avatars and Mirage-powered creation, while new creator stories show faster end-to-end production.

13 · Localization

Localization is recurring, but language counts are sparse

HubSpot

The case says HubSpot uses native multilingual generation to make variants across markets without added cost or complexity. No exact language count is published.

Plug

Plug’s workflow explicitly uses localized AI-generated ad content for e-commerce clients. The case reports reduced localization cost but no language count.

Evidence rule: “multilingual” is not converted into a numerical language total. Missing counts remain “Not publicly disclosed.”
14 · Negative evidence & limitations

What the public record does not prove

No audited ROI

Public case studies do not provide enough investment and counterfactual data to independently calculate customer ROI.

Few independent validations

The strongest quantified metrics are mostly published by Captions/Mirage in customer stories rather than independent audits.

Limited enterprise detail

Some enterprise users are named only in corporate announcements; detailed workflows and metrics are not public.

Attribution is difficult

Views, conversion and revenue can change for many reasons. Customer attribution is not a controlled experiment.

Product drift

Historical stories describe earlier capabilities and should not be treated as proof of identical 2026 functionality.

Demo ≠ customer

Current AI UGC, avatar and Mirage examples demonstrate capabilities, but are not automatically commercial deployments.

15 · Competitive context

Workflow context, not competitor ranking

The reviewed case studies are primarily about workflow transformation rather than head-to-head tool comparisons. Customers describe compressing manual production, creator sourcing, editing and localization. Captions’ current editorial material discusses tools such as CapCut, but that is not customer evidence.

CapCutAdobeDescriptOpusClipRunwayHeyGenSynthesiaCanvaVEED

No competitor usage is assigned to any customer unless verified.

16 · Evidence scoring

Case-study quality, not “best customer” ranking

The prompt requested a 100-point case-study quality score. Here it is used as an evidence lens rather than a customer-success ranking.

Plug

Source quality19/20
Quantitative evidence19/20
Overall evidence lens88/100

Peter

Source quality19/20
Quantitative evidence15/20
Overall evidence lens82/100

HubSpot

Source quality20/20
Quantitative evidence12/20
Overall evidence lens84/100

These scores are a research heuristic about evidence quality, not a ranking of customer success or product value.

17 · Key insights

Evidence-based patterns

1. Production speed is the recurring bottleneck

Public cases disclose four hours → under one hour, hours → 15 minutes, one month → one day, and nearly three weeks → one day for campaign launch.

2. UGC is the clearest enterprise marketing use case

HubSpot, Plug and the anonymous startup use Captions for creator-style advertising and performance creative.

3. Localization is an amplifier

Multilingual generation is repeatedly described as a way to expand campaigns without repeating traditional production. Public cases are qualitative on language count.

4. Creator value is consistency

Bonsa and Elise show that less finishing work can make regular publishing more sustainable.

5. Performance claims need attribution discipline

CTR, conversion, CPL and views are reported in customer stories, but they are not independent causal proof.

6. Evidence gaps matter

Subscription costs, experimental controls, independent audits and detailed adoption data are usually missing, limiting rigorous ROI analysis.

18 · Research conclusion

What the public record supports

Evidence-led conclusion

Captions’ documented cases show a consistent pattern: AI is used to compress video-production work — especially editing, creator-style ad generation, variation, captioning and localization — so small teams can publish or test more creative with less manual production.

The strongest quantified stories are Plug’s one-month-to-one-day ad-delivery turnaround, 5× lower CPL and 50% lower production costs; the anonymous startup’s +83% CTR, +33% conversion rate and +41% hook rate; Peter’s under-one-hour finished videos and more-than-doubled views; and Bonsa’s 15-minute finishing workflow.

However, the public evidence is predominantly first-party and customer-reported. It does not support an independently calculated customer ROI benchmark, a universal causal claim that Captions caused revenue growth, or a complete enterprise customer list.

19 · Visual assets

Publication imagery

Gaurav Misra — supplied image
Supplied Gaurav Misra image used as the editorial portrait.
Captions logo — supplied image
Supplied Captions brand image used as the visual product identifier.
20 · Source audit

Primary sources used

Mirage — AboutCorporate identity, Captions app, foundation models, API.Open source ↗
Introducing Mirage — 4 Sep 2025Captions → Mirage rebrand; Captions app remains central.Open source ↗
HubSpot — 28 Oct 2025UGC workflow, variations, multilingual generation and CMO quote.Open source ↗
Peter — 21 May 2026Four-hour old workflow, under-one-hour result, Prism Pro, views and listing opportunity.Open source ↗
Bonsa — 30 Jun 202615-minute finishing time and nearly $100K generated through content.Open source ↗
Anonymous startup — 2 Sep 20252+ weeks/video, one-day launch, +83% CTR, +33% CVR, +41% hook rate.Open source ↗
Plug — 9 Dec 2024AI Creators / AGC, one-month-to-one-day delivery, 5× lower CPL, 50% lower production cost.Open source ↗
Elise Micheals — 22 Aug 2022Historical captioning workflow, five-minute process and engagement report.Open source ↗
Mirage financing — 24 Mar 202620M users, 250M videos and enterprise-user mentions including HubSpot, CoreWeave and King.Open source ↗
Captions Help — IntroductionCurrent product capabilities including AI Edit, avatars, dubbing and generation.Open source ↗
Mirage Studio / Ad Studio documentationCurrent generated-actor and AI-ad capabilities; product evidence, not customer proof.Mirage Studio ↗ · Ad Studio ↗
AI Edit guide — updated 2026Current AI Edit behavior: transitions, subtitles, visuals and music.Open source ↗
AI dubbing guide — 22 Jul 2026Current localization/dubbing workflow.Open source ↗
Publication note: This report is designed for Artificial Intelligence Files. Numerical claims are tied to cited sources; company-reported and customer-reported outcomes are explicitly labeled. Missing information is intentionally left undisclosed rather than inferred.