Headline & intro
AI video startups are no longer competing on filters and flashy demos, but on something more fundamental: who can encode the instincts of a good editor into a model. Mirage – the company behind the Captions app – just raised $75 million to do exactly that. This isn’t just another round in the noisy "AI for creators" space. It’s a signal that investors believe there is defensible IP in models that understand pacing, framing, attention, and even accents. In this piece, we’ll unpack why Mirage’s shift from app to "AI lab" matters, how it fits into the battle against CapCut and Canva, and what the move tells us about the next phase of AI video – especially for European creators and SMEs.
The news in brief
According to TechCrunch, Mirage, maker of the AI video editing app Captions, has secured $75 million in growth financing from General Catalyst’s Customer Value Fund.
Over the last year, the company has rebranded from Captions to Mirage, positioning itself as an AI research and model-development lab rather than just an app vendor. It has built a dedicated model for short-form video that optimises pacing, framing and where viewers’ attention should go.
Mirage also launched a freemium business model in early 2025 to compete more directly with ByteDance’s CapCut and Meta’s Edits. Alongside the mobile-first Captions editor, it now offers a web-based video creation suite aimed at companies producing marketing content at scale, with plans to merge the two.
AppFigures data cited by TechCrunch indicates over 3.2 million downloads and around $28.4 million in in‑app revenue in the past 12 months. The company says users have generated more than 200 million videos, and that only about a quarter of its revenue comes from the U.S., pointing to a heavily international user base. Mirage plans to use the new funding to build more models and expand in fast‑growing Asian markets.
Why this matters
Mirage is interesting not because it makes AI videos – so do dozens of rivals – but because it is quietly attacking one of the hardest problems in media: editing judgement.
Most AI video tools today either:
- generate clips from text prompts, or
- help with low-level tasks (captions, background removal, translations).
Mirage is betting that the real value lies in models that understand how to assemble a compelling short video from disparate inputs: different takes, B‑roll, audio tracks, overlays. This is what the company calls "assembly intelligence". In practice, that means learning the grammar of TikTok, Reels and Shorts: when to cut, where to crop, when to insert a zoom or meme, how long to stay on a face.
If Mirage can codify this into models, several consequences follow:
- Creators get leverage, not just automation. Instead of replacing creators, such tools amplify those who already know their audience, by offloading the tedious but crucial micro-decisions.
- Brands and SMEs get something closer to an in‑house agency. Small teams that can’t afford professional editors suddenly have access to workflows that mimic them.
- Mirage becomes harder to copy. UI can be cloned, but a model trained on millions of real edits and performance outcomes is an asset with compounding advantage, especially if it is fed by continuous use.
The accent-preserving audio model is another underappreciated move. For an international user base where three‑quarters of revenue is non‑US, sounding authentically Brazilian, Indian, Nigerian or Spanish isn’t a nice-to-have – it’s core to trust and engagement. This is an area where Western competitors often underinvest, and where Mirage can differentiate strongly in emerging markets.
The bigger picture
Mirage’s round lands in a crowded field. TechCrunch notes that Canva, D‑ID, HeyGen, Webflow and Avataar are all pushing AI video pipelines for marketing. Add to that the generative giants like Runway, Pika, Stability, and the built‑in tools in TikTok, Meta and YouTube. On the surface, this looks like yet another player chasing the same "AI video for marketers" TAM.
Look closer, and two deeper trends emerge:
Verticalisation of AI tooling. We’re moving away from generic "edit anything" apps towards stacks tuned to specific formats and workflows. Mirage has picked its battlefield: short-form, mobile-first, social-native content and performance marketing. That clarity is attractive to investors because it narrows product scope while keeping a huge market.
From SaaS features to model-native companies. Many incumbents bolt AI features onto existing editors. Mirage has flipped it: the company is reorganising itself around building proprietary models first, and apps second. That mirrors what we’ve seen with Figma’s acquisition moves, Adobe’s Firefly push, and Runway’s platform strategy – whoever owns the underlying models controls distribution and margins.
There is also a historical parallel. A decade ago, YouTube MCNs and creator tools tried to productise "growth hacking" knowledge. Most failed because growth playbooks changed constantly. AI now offers a way to continuously relearn those patterns at scale by watching what users actually do and how audiences react.
If Mirage executes, it could become to short-form performance video what Shopify became to ecommerce: an enabling layer that abstracts complexity for millions of small businesses, while quietly aggregating behavioural data that keeps improving the system.
The European / regional angle
For European creators and SMEs, Mirage’s trajectory raises both opportunities and flags.
On the opportunity side, Europe is full of small brands that live or die on Instagram and TikTok, but can’t afford creative agencies in London or Berlin. Tools that can auto‑edit, localise and version hundreds of short clips – with correct languages and accents – are a direct productivity boost. For a Slovenian boutique, a Croatian hotel chain or a German D2C brand, that’s the difference between posting twice a week and running always‑on campaigns across multiple markets.
But the regulatory environment is very different from the Asian expansion Mirage is prioritising. Any serious European play will run into:
- GDPR and data residency. Training and improving "assembly" models requires ingesting large volumes of user content and performance data. European regulators will expect clear consent, purpose limitation and possibly localisation of some datasets.
- The Digital Services Act (DSA). As AI‑assisted content floods platforms, transparency around sponsored content, political messaging and deepfakes becomes critical. Vendors like Mirage will face pressure to build watermarking, disclosure and safety controls into their tools.
- The EU AI Act. Depending on how models are marketed (e.g. for political campaigning or employment-related content), they could fall into higher-risk categories with stricter requirements.
There is also local competition. European-born companies like Synthesia (UK), Colossyan (originally Central Europe), and various Berlin and Paris‑based video startups are already entrenched in enterprise and marketing use cases, often with privacy and compliance as selling points. Mirage entering this arena would force all sides to clarify their stance on data use and content authenticity.
Looking ahead
Several things are worth watching over the next 12–24 months.
Can Mirage truly merge creator and enterprise products? Building a mobile-first editor for influencers and a web-based suite for marketing teams are very different challenges: UX expectations, pricing, support and security requirements diverge quickly. If Mirage actually unifies these into one coherent platform, that’s a moat. If it ends up as two loosely coupled products, incumbents can outflank them.
Will "assembly intelligence" stay proprietary? If Mirage’s models prove effective, two pressures will mount: platforms like TikTok and Meta will want similar capabilities built‑in, and open‑source communities will try to replicate them. Mirage’s defence will likely be a combination of data scale, closed feedback loops and partnerships.
How aggressive will regulators become on synthetic media? Deepfake scandals in politics or advertising could trigger stricter rules on disclosure and labelling. Any company optimising engagement with AI-
assembled video might suddenly be required to surface how much of a clip is synthetic, which assets were auto-generated, and what data they were trained on.Monetisation and margins. Freemium is necessary to compete with CapCut, but compute costs for heavy AI models are brutal. The investor quoted by TechCrunch emphasised strong unit economics; the key question is whether Mirage can maintain that as usage scales and as users expect near-real-time generation on mobile.
My expectation: the market will consolidate. A few global players will own the end-to-end stacks for short-form video marketing, while many others become niche plug‑ins or get acquired. Mirage is clearly positioning itself to be one of those few, not a feature.
The bottom line
Mirage’s $75 million round is not just fuel for another AI video app; it’s a wager that editing judgement can be learned, scaled and sold as a service. If the company can turn "assembly intelligence" and accent-aware audio into reliable, compliant, high-margin products, it will sit at a powerful intersection of creators, SMEs and big brands. If not, it risks being squeezed between free tools from TikTok/Meta and compliance‑first European rivals. The open question for readers: would you trust an AI not just to cut your video, but to decide what your audience should see and hear?
