Two Takes On How AI And Video Are Intersecting In Fashion

As we saw from the findings of 2026 AI in fashion survey (as distilled into the Arm’s Length AI essay and agent toolkit, and as captured in full, granular detail in the AI Report 2026) fashion, on aggregate, sees visual content generation one of the most mature use cases for AI today, as well as predicting that an increased amount of customer-facing content will be either AI-generated or AI-enhanced in the coming year.

This optimism covers both static images and video, both of which are seeing extensive use in creative campaigns and on everyday eCommerce product detail pages. And while image workflows have been the easiest for fashion – along with other industries – to grasp, image models have lately stalled in terms of pure capability and output progress, while video generation models have continued to make more pronounced leaps.

The upshot of which is that both flat image and video content on brand and retailer websites is now far more likely to have been enhanced or generated by AI than consumers expect. Labelling regulations will change this perception, of course, but the introduction of mandatory disclosure for AI-created content will also serve to spotlight just how embedded generative workflows have become in the business of both inward and outward-facing visual content, moving or otherwise.

And this week’s headlines provide even further evidence that the expectations our survey participants have, for even greater use of generative models in visualisation and marketing, are likely to be well-founded.

First, Higgsfield, the video and image platform that’s become a household name in AI content creation (and also a lightning rod for the creative criticism that invites, naturally) announced the completion of a $400 million funding round, which values the company at nearly $5.5 billion USD. 

That valuation is, as Reuters puts it, predicated on fast-scaling demand for AI-generated marketing and media content. But while we’d have needed to caveat that with an “expected” qualifier fairly recently, the revenue claims made as part of the announcement ($700 million annualised, i.e. based on the most successful month in recent history, and extrapolated over a twelve month period) suggests that said demand is already very real.

And while Higgsfield has made a major play for moviemaking through the studio system and the hobbyists (with the caveat that movies like Obsession prove just how blurry the lines between those ecosystems are), the company is also one of several cross-industry platforms to stake a big claim to fashion’s brand-native and UGC content market, with its Soul post-trained model and other tools.

There is, in other words, a lot of money flowing into video generation for fashion. And that’s before we extend our scope to include the other announcement from this week, that Anthropic (the frontier lab behind the Claude applications and model family) are apparently, according to Bloomberg, in talks to buy an AI startup called Decart for $6 billion.

Decart might not be a name a lot of our readers are familiar with, but they may have seen some of the demos of the company’s Lucy real-time video model, which recently saw a milestone release that improved its ability to ingest a video feed and, with a perceptible but small delay, stream a generated result that could swap out elements in the original feed, which included, as an obvious use case, something approaching real-time virtual try-on. The Interline had a closed doors preview of the previous generation of Lucy, and while the visual approximation of fabric simulation was definitionally not perfect, it was more than sufficient for the intended purpose.

This is, to be very clear, a fundamentally different proposition to body project mapping. The proposition with Lucy (and we should point out that Decart does have other products and business units) is not taking a 3D garment model and then capturing crude skeletal anchor points from a video feed and mapping the 3D model to it. It is, in a very literal sense, taking one or more reference images of a garment, along with a real-time running video feed from a camera, and then generating a new video output that combines those things. And the results, at least in our limited exposure to the prior generation, were impressive.

This headline appears to have flown under the radar this week, but when taken at simplistic face value as “company behind Claude is seemingly ready to pay billions for real-time VTO,” it should give all our readers pause. Not just because this appears to be an indication that there is still much more to come in pure model maturity, but because it suggests a coming wave of ubiquity that could make the current ethical and cultural concerns around generated content look comparatively chill.

But, as it happens, video generation is just one of two areas where AI is making headlines in fashion this week, and the stakes are equally high in the opposite camp as well. 

This other avenue lies in the application of AI on top of incoming video feeds, or prerecorded footage, for analytical rather than generative purposes. And the most directly relevant for fashion was the launch of GSD Realmotion, which is, arguably, the most profound vote of confidence we’ve yet seen in upstream AI.

This newsletter is not the place to rehearse the entire history of motion / time / method study as a means to calculate standard minute values, but suffice it to say that, as a scientific foundation for costing, negotiation, labour arbitrage and so on, GSD’s methods (approved by the International Labour Organisation, no less) occupy a high pedestal for accuracy and accountability. The standard may not always be used as readily as it probably should be, since brands seemingly prefer to rely on historical averages and so-called ‘napkin math’ in negotiations, but the promise of Realmotion – that an AI vision model can examine smartphone footage of factory operations and then either assign them pre-existing codes or use them to potentially generate new ones – represents a marked step up in the industry’s potential trust in AI.

And although the use case is very different, the same technology principle of applying AI vision models to existing video footage – in this case CCTV – was seemingly behind a very divergent set of outcomes for loss / shrinkage protection in retail this week.

In the first story, Iceland (a UK supermarket known for its affordable frozen food, but which also sells other food and beverage categories) announced that it had reduced theft-related losses in its stores by 80% after applying AI on top of the video feeds from its existing in-store camera networks. 

Notably, this approach seemingly did not involve trying to detect and understand people, either in motion or through personally identifiable information, and instead relied on monitoring stock on shelves, through vision rather than by physically tagging the goods – presumably because the latter approach would have been prohibitively expensive for low-value food items.

But nevertheless, Iceland, in the same announcement, also floated the idea of adopting face-detection systems that it could layer on top of the same existing security camera estate, and that would cross-reference incoming customers against a pre-existing roster of known shoplifters to identify theft risks.

This expansion, though, may not be something fashion retailers (or Iceland itself) want to consider, since the news headlines this week also did the high-stakes-finding for us. The day prior to Iceland’s announcement, larger UK supermarket chain Sainsbury’s – which notably does sell own-brand fashion under the Tui label – hit the news because its implementation of exactly that kind of face-detection technology allegedly led to security guards escorting an innocent man off the premises after his face was incorrectly matched to the risk register.

There’s no reason, at least for the moment, to assume that there is anything in common between the AI vision models being deployed upstream, in factories, and downstream, in retail settings, but the risks of getting things wrong in either context are potentially severe at a time when scrutiny of fashion brands and retailers, across the board, is so high.