How AI Is Actually Being Used in Product Photography Right Now?

Unilever reported that its AI-driven product shoot pipeline delivered up to 55% cost savings, 65% faster turnaround, roughly three times the dwell time, and double the click-through rate on the assets it produced. That is the honest headline number from one of the largest FMCG groups in the world, and it did not come from typing a prompt and hitting generate. It came from a hybrid workflow that sits somewhere between a CGI studio and a traditional shoot.

That is the story most takes are still missing. The discourse around AI in product photography keeps bouncing between "studios are over" and "the images are obviously fake". Neither reflects what brands are actually doing in 2026.

This is a working map of where AI genuinely sits in the product photography pipeline right now: what it does well, where it still falls over, and what that means for the people writing the briefs.

What "AI in product photography" actually covers in 2026

The phrase has been stretched to mean four different things, and most arguments about it are really arguments about which one someone is picturing.

The first is full generation from a text prompt. A bottle, a can, a pair of sneakers, summoned from nothing. This is what gets the headlines and the viral tweets. It is also the part of the stack that matters least for serious brand work.

The second is image-to-image on a real product plate: you shoot the product, then use AI to restyle the environment, swap backgrounds, or extend the frame. The third is AI-assisted retouching and clean-up inside a traditional post-production pipeline. The fourth is 3D digital twins feeding AI-driven variant generation, which is where the big CPG groups have actually been spending their budgets.

Most of the real work is happening in categories two, three, and four. Category one gets most of the attention.

Where AI is genuinely earning its place

The clearest wins are in the long tail of product imagery: the thousands of marketplace listings, seasonal variants, size and colour permutations, and localised ad creatives that a brand needs on an ongoing basis.

ASOS, by published accounts, hit a 73% reduction in studio photography costs within six months of deploying AI-generated lifestyle imagery across more than 850 clothing lines in early 2025. H&M's digital team reported similar savings while increasing catalog output in the same period. These numbers only make sense once you accept that the bulk of e-commerce imagery was never going to justify a full studio day in the first place.

Nestlé has gone further by building digital twins of its products and reported cutting advertising time and cost by around 70% through that approach; teams swap backgrounds, pack copy, and market-specific details without reshooting. Coca-Cola, working with Adobe on Project Fizzion, has been training brand rules into a design system that scales localised ad variants across hundreds of markets.

There is a cost-structure reason for all of this. Traditional product shoots typically run $200-$5,000 or more per session, depending on the product, the crew, and the requirement. AI-generated imagery, at the unit level, can run between roughly $0.10 and $2.00 per image. For the long tail, that gap is unarguable.

A BigCommerce analysis of around 12,000 online stores in late 2024 and early 2025 also found that merchants who moved to AI-enhanced product imagery saw conversion uplifts between 35% and 67%, with a median of 49%. Part of that is the lift from doing any imagery upgrade at all, but the speed and volume are what make it feasible for smaller catalogs.

Beyond volume work, AI has also quietly become standard in pre-production. Concept frames, mood boards, and "let's see what this idea looks like before we build it" mockups now take minutes instead of a day. That is a real productivity gain even for studios that end up shooting the final frame on camera.

Where it still falls over

The failures are where you would expect them: the moments where the product has to be the product.

Reflective surfaces are the first tell. Glass, liquid, metal, and any curved chrome detail still come out subtly wrong in pure AI generation. Highlights sit in the wrong places, internal reflections contradict the environment, and the hand-off between product and background often reads like a cut-out. For a shampoo bottle on a retail listing, that is survivable; for a premium fragrance hero frame, it is not.

Material texture is the second. Fabric weave, grain on leather, brushed aluminium, ceramic glaze — AI gets close, but close costs money in returns and customer trust when the product in the box looks different from the one on the page.

The third is brand-critical detail: logos, label typography, regulatory pack copy, exact colourways. These are not subjective. A generated render that gets the Pantone off by a few units, or reflows the ingredient list, does not ship.

There is also the compliance layer. Amazon and most European marketplaces have strict rules for PDP hero images — clean background, no overlays, accurate representation of the product. Most brands still start from a real plate for those frames, and then use AI only to clean, extend, or replace the surround.

This is why the dominant pattern in 2026 is hybrid, not replacement. The big savings numbers above come from pipelines where traditional photography still owns the hero, and AI owns the variant, the scale, and the clean-up.

What this looks like inside a real post-production pipeline

Inside our own studio, the division of labour has settled into a pattern that most working post houses would recognise.

At the front of the pipe, AI does concept work. We will generate half a dozen directions on a packaging idea or a campaign frame in an afternoon, kill most of them, and take the surviving two into real production. That used to be a week of illustration and moodboarding; it is now a morning.

Mid-pipe, AI shows up in clean-up, extension, and variant work. If we have a strong studio plate of a bottle but the client needs ten market-specific backgrounds, we are not reshooting — we are building variants off the master. The highlights on the bottle, though, still get rebuilt in compositing by hand when the plate has a blown reflection or a dust mark. That is the quiet daily craft work that determines whether the final image holds up on a billboard.

On hero frames, packaging design, and anything where the product has to read exactly right, we stay on traditional craft: photography, CGI, or hybrid photo-plus-CGI composites. The hybrid route — a real product plate combined with a CGI pour or a 3D background — is increasingly the sensible middle path for brand work. It gives the accuracy of a shoot with the flexibility of a render.

AI is a layer in that pipeline, not a substitute for it.

What this means for the brief you write

For agency creatives and brand managers commissioning work in 2026, the practical question has shifted. It is no longer "AI or not" — it is "where in this job does AI belong".

A few questions worth asking whoever is quoting the work:

• What is getting shot, what is getting generated, and what is getting composited?
• Is there a master plate or a digital twin the variants will come off, or is each image being generated independently?
• Who is checking material accuracy, brand colour, and pack copy before delivery?
• What is the licensing position on any AI-generated elements, and how is the studio documenting it?

The brands getting the real savings are the ones running governed pipelines: approved material libraries, locked-down brand rules, version control on digital twins. The ones getting burned are the ones treating AI as a free image tap.

For smaller DTC brands and e-commerce teams, the calculus is different but the principle is the same. AI is fine — usually better than fine — for listing, variant, and social work. For the hero image that defines how the brand looks, the economics of spending a little more on craft still hold.

The short version

AI is real in product photography, and it is no longer a side experiment. It is cutting time and cost at the long-tail end of the pipeline, and it is freeing up budget for the frames that still need a human eye.

What it has not done is collapse the work into a single prompt. The studios and brands getting the best results are running hybrid pipelines with discipline on both sides of the seam. Everything else is still hype.

If you are working on a project that sits in this middle ground — hero frames plus a long tail of variants, or a packaging build that needs both CGI and shot elements — we are always happy to talk.