Nick Knight on AI as a New Creative Medium and Why Fashion Should Stop Looking Backward

AI is enabling a genuinely new artistic medium—moving still images—that exists between photography and film, but most creators waste the technology by simply mimicking existing 20th-century work.

Summary of Nick Knight Is Building His Own AI. Every timestamp links into the original audio.

The short version

• 00:00:34 — Knight is building AI datasets from scratch by repeatedly photographing models so the machine learns only from his specific visual references rather than scraping the internet for generic training data.
• 00:00:49 — He takes still photographs and uses AI to animate them, creating something that is neither traditional photography nor film—a new hybrid medium that maintains the original image’s emotional weight and lighting while adding movement.
• 00:04:36 — The metaverse hasn’t disappeared; the term was abandoned due to negative associations, but the underlying concepts of virtual worlds, avatars, and digital identity remain active even if fashion’s interest has waned.
• 00:17:17 — Knight uses a mix of open-source tools and custom in-house AI systems at Show Studio, constantly experimenting with what’s possible rather than defaulting to pre-existing software.
• 00:20:09 — To avoid AI simply mimicking existing artists, Knight creates isolated datasets where the AI only has access to his own photographic references, preventing the system from defaulting to famous photographers’ styles.
• 00:26:46 — Fashion models should own their own digital likenesses and control how their virtual versions are used, yet model agencies currently lack incentive to invest in protecting these rights because they profit from the current system.
• 00:28:52 — Young image makers should abandon trying to recreate 20th-century photography or chase magazine work, instead pursuing what genuinely excites them and provides fulfillment rather than chasing fame, money, or power.
• 00:35:16 — Knight’s fulfillment comes from human interaction—with staff, family, colleagues, and strangers—rather than from creating trophy pieces or achieving recognition, and he approaches his work as ongoing conversation rather than finished masterpieces.

In depth

What actually happened to the metaverse

Ahmed opens by pressing Knight on a claim he made back in 2022, when the metaverse was the industry’s favourite buzzword and Knight himself speculated that physical runway shows might eventually become unnecessary 00:04:18. Three years on, the word has all but vanished from fashion conversation, and Ahmed wants to know whether Knight was simply wrong. Knight’s answer is a hedge rather than a retraction: he argues the concept never actually disappeared, only the label did, and he pins the blame partly on Meta’s decision to brand itself around the term, which he thinks poisoned the word for a wide swath of people 00:04:47. Avatars, NFTs, and virtual identity, he insists, are all still humming along in the background even though nobody bothers to call them ’the metaverse’ anymore 00:04:55.

What’s more interesting than the semantic point is Knight’s diagnosis of why fashion specifically lost interest. He distinguishes between people who are genuinely nostalgic for pre-digital ways of experiencing clothes and imagery — the tactile, magazine-era sensibility — and people who have a straightforwardly commercial reason to defend the status quo: their income, status, or job security depends on things not changing 00:06:15. He’s careful not to dismiss the nostalgic camp as wrong, calling their attachment ‘a true feeling’ 00:06:12, but he frames both groups as short-termist, unable to see past a 20th-century frame of reference that he says is actively distorting how the industry reads the 21st century 00:05:45.

Ahmed doesn’t fully let this go, pointing out later that what has survived from the Plato’s Atlantis era of live-streamed shows looks less like open creative experimentation and more like highly produced advertising aimed at driving attention and sales 00:13:14. Knight doesn’t really contest the observation; instead he sidesteps it, saying the question doesn’t sit well with how he actually experiences the world 00:13:26, and pivots into a broader argument about fame and commercial incentives distorting judgment — effectively conceding that fashion’s current use of live technology is money-driven without defending it on those terms.

Building a clean dataset and a medium with no name

The most concrete, hands-on part of the conversation concerns what Knight is actually doing inside Show Studio right now. He describes deliberately building his own AI training data by shooting a model repeatedly and exhaustively across a full fashion session, so that when he asks a system to generate a sculpture, a poem, or a moving image from that material, the machine has nothing else to draw on but his own photographs 00:21:14. The stated purpose is to sidestep the most common criticism of generative tools: that they default to visual clichés lifted wholesale from famous photographers, spitting out something that looks like ‘a Stephen Meisel picture or a Mert and Marcus’ the moment you type in a vague prompt 00:19:55. By isolating the reference pool, Knight is trying to guarantee that whatever comes out is at minimum traceable to material he made and owns, even if the AI is doing the interpretive work.

From that dataset, he describes a genuinely new working process: take a single still photograph and ask the AI to animate it, producing a result that inherits the original picture’s lighting, colour balance, and emotional tone but now moves — not video, because it wasn’t built from a sequence of frames, and not a photograph anymore either 00:18:08. He argues this is a legitimately new artistic state that doesn’t yet have a name, and that the creative labour has simply relocated: instead of composing a shot, the artist is now composing the prompt and the iterative back-and-forth with the model that shapes how that stillness becomes motion 00:18:35. He extends the idea further — turning a 2D image into a three-dimensional sculpture, printing that sculpture, painting it, rephotographing it, then having the whole object rendered back out as a poem — describing this chain as a creative journey of testing rather than a shortcut to a finished product 00:19:19.

Crucially, Knight frames the interesting failure of the current moment not as AI itself but as unimaginative use of it: most people, he says, are using extraordinarily novel tools to painstakingly recreate old aesthetics, an Avedon-style shoot or a Bruce Weber mood, rather than pushing into whatever this new hybrid form could become 00:19:40. That’s presented as a problem of nerve and imagination among image-makers, not a limitation of the technology.

Referencing, plagiarism, and whose data is it anyway

Ahmed pushes Knight on the ethics of AI systems trained on other artists’ work, and Knight’s response is less a defence of AI than an attempt to complicate the premise of the question. He argues that all human creativity works by reference — that walking through a museum deposits Botticellis and mid-century furniture into your subconscious, which then surfaces unbidden when you make something new — and that AI models function analogously, drawing on whatever visual corpus they’ve absorbed 00:20:33. He backs this with an anecdote about the advertising industry’s actual practice: nearly every commissioned fashion or beauty campaign arrives with a mood board plastered with uncredited, unpaid images from Meisel, Mert and Marcus, or Knight himself, and nobody in that supply chain treats it as theft 00:22:27. His conclusion is pointedly ambivalent — he says he’d rather be part of a shared visual vocabulary than excluded from it 00:22:48 — but he’s careful to flag, twice, that he isn’t volunteering to be AI’s defender, just offering a perspective that complicates a simple ‘it’s plagiarism’ framing 00:23:19.

Where he does draw a firm line is on quality of outcome rather than legality or ethics. Knight insists that mimicry is intrinsically unsatisfying — that copying an existing image or photographer’s style yields no creative reward regardless of the tool used — and that this is precisely why great AI-assisted work still requires the same obsessive, time-consuming iteration that analogue printing always demanded, invoking his own history spending three days in the darkroom perfecting a single Jil Sander print with Brian Dowling 00:24:12. In his account, the tool has changed but the psychological and technical labour of getting something right hasn’t shrunk at all.

This leads into his own workaround: rather than adjudicate the ethics of scraped datasets in the abstract, he simply refuses to use them, building closed, self-generated datasets so the AI has nothing borrowed to draw from 00:17:21. It’s a practical, almost craftsman’s answer to a legal and philosophical debate he otherwise treats as murky and unresolved — he never claims his approach settles the wider industry argument, only that it lets him personally avoid the accusation.

Who owns a model’s digital face

When Ahmed narrows the referencing debate to a specific and thorny case — companies licensing or replicating a real model’s likeness — Knight’s position hardens considerably and loses its earlier ambivalence. He starts from an observation about the structural unfairness of modelling as a career: it has traditionally offered a brutally short working life because the industry is fixated on youth, favoring people in their late teens and early twenties 00:26:01. He frames AI-driven digital likeness as a potential correction to that unfairness, since it could in principle let a model’s image keep working — and keep earning for them — long after their body has aged out of demand 00:26:15.

But he is explicit that this potential benefit is currently being captured by nobody, or worse, by everybody except the model. As things stand, anyone can generate a virtual version of a known model without consent, payment, or even notification, and the model has no legal claim over that likeness 00:26:40. Knight’s proposed fix is unambiguous: models need enforceable ownership of their own digital selves, with the ability to control which version of themselves — eighteen years old or forty — gets deployed and where 00:27:32.

He’s notably pessimistic, though, about who will actually build that protection. Having spoken with agencies directly, he reports that there is essentially no movement toward securing these rights, and he attributes the inertia to a straightforward incentive problem: agencies already have a profitable, low-cost business model built on cycling in new young talent, so there is no commercial reason for them to spend the money required to create and safeguard digital likenesses for models they already represent 00:26:52. He layers onto this a broader, unresolved indictment of the modelling industry’s history of mismanagement, careful to note that plenty of agents do right by their clients, while insisting the systemic protections needed for the AI era simply don’t exist yet 00:27:50.

Fulfillment over fame, and advice with no rulebook

The conversation’s emotional center is Knight’s repeated, almost irritated rejection of fame, money, and power as legitimate goals — a theme Ahmed keeps circling back to because Knight keeps volunteering it unprompted, even when the question was ostensibly about live-streaming or red carpets. Knight argues bluntly that he has met people who are rich, famous, and powerful, and none of them report being happier for it, some visibly less so 00:11:57. He extends this into a mild critique of the entire attention economy fashion now runs on, saying he doesn’t understand the psychology of screaming at a celebrity’s car while ignoring the fact that a stranger at a bus stop is just as interesting if you’re willing to actually engage with them 00:14:18.

That philosophy is what shapes his answer when Ahmed finally asks the practical question every young reader wants answered: what should someone starting out in image-making today actually do? Knight’s advice is deliberately unspecific to the point of refusing to give a technical roadmap at all — he compares trying to give concrete guidance in this moment to standing in a field advising bystanders on career planning as the first steam locomotive rattles past, arguing that the pace of change makes any tactical advice obsolete almost immediately 00:29:14. What he offers instead is a stance: stop measuring yourself against 20th-century benchmarks like magazine work, which he considers functionally irrelevant now, and instead orient toward whatever genuinely produces a feeling of aliveness in you, regardless of medium or discipline 00:29:50.

Notably, Knight distinguishes this ‘fulfillment’ explicitly from happiness, which he calls a kind of delirium not worth chasing on its own — he’d rather have the full range of human emotion, sadness and fear included, than flatten life into constant contentment 00:34:32. His own account of where fulfillment comes from is unglamorous: greeting his studio assistants each morning, ordinary interactions with strangers, his family — not the accumulation of acclaimed images 00:35:37. He goes so far as to reject the idea of a ‘best’ photograph altogether, treating his body of work as an ongoing conversation rather than a trophy case, and answering the inevitable ‘what’s your best shot’ question by saying it’s always the next one, since he doesn’t yet know what it will be 00:36:34.

Summarised automatically. Listen to the original for the full conversation — this is not a substitute for it.