Kaedim.

A Machine, a Room of Artists, and the Gap Between Them

Here is a fact about video games that sounds made up but isn't: the average budget for a big-studio title went from roughly $10 million in 2010 to well north of $200 million today. That's a 20x increase in about fifteen years, and most of it isn't marketing or executive salaries. It's content - the swords, the buildings, the rocks, the characters, the thousands of individual 3D objects that a modern game needs to feel real. Someone has to model all of that. Traditionally, that someone is a human artist, working slowly, at a price that scales linearly with the size of the world.

Kaedim is a company built on the premise that this is a solvable problem. Its product does something that, described plainly, sounds like a magic trick: you feed it a 2D image - a piece of concept art, a photo, a sketch - and it hands you back a 3D model, ready to drop into a game engine. The pitch to studios is speed and cost: assets up to 10 times faster, at up to 90% less than the traditional outsourcing bill. When your content costs are growing 20x, a tool that cuts one of them dramatically is not a nice-to-have. It's a line item the CFO actually wants to hear about.

The company was founded in 2020 by Konstantina Psoma and Roman Bromidge, who met as computer science students at the University of Bristol. Psoma, the CEO, grew up in Athens - and the detail everyone who writes about her keeps, because it's genuinely useful, is that she grew up watching her father build a software company and her mother make art. Kaedim is a business that requires being fluent in both. She finished a masters in deep learning, and then did the thing more founders should do: before writing much code, she talked to actual game developers about what actually hurt. Some of them - the UK studios Rebellion and Aardman among them - liked the answer enough to become her first backers.

The interesting part is what it isn't

The temptation, in 2023 and 2024, was for every company touching 3D generation to describe itself as pure AI - press a button, receive a model, no humans involved. Kaedim spent a period of its early life leaning into exactly that framing, and it caused the company real trouble. In September 2023, the outlet 404 Media reported that a meaningful amount of the "AI" output was actually being produced by human contractors around the world, some paid only a few dollars per model, in roles Kaedim described as "quality control." Some of those workers said they were building models from scratch. It was, briefly, a case study in the gap between how an AI product is marketed and how it actually runs.

What Kaedim did next is the genuinely instructive bit. Within hours of the report, it changed its own messaging - not to deny the humans, but to put them on the label. The site began describing the product accurately: machine learning and an in-house art team, combined, delivering production-quality assets. This is now the company's stated model, and it has a name that has since become fashionable across the whole AI industry: human-in-the-loop. The machine drafts. A trained artist refines. The output goes out the door as something a studio can actually ship.

You can read that pivot cynically - a company caught overstating its automation and forced to admit it. You can also read it as the more durable version of the business. A fully automated 3D generator that produces almost-right models is a demo. A pipeline that reliably produces game-ready assets, because a human catches the things the model gets wrong, is a product studios will pay for on a Tuesday. The second thing is harder to build and easier to sell. Kaedim, somewhat by force, ended up describing the second thing.

What you can actually do with it

In practice, Kaedim is a web platform with a credit-based subscription behind it. An indie developer prototyping a game on a weekend can buy a small pack of credits and turn their concept art into models. A large studio signs a custom enterprise contract. The company reports serving over 20,000 new creators a month alongside 250-plus enterprise developers - a spread that runs from solo hobbyists to AAA studios, film production houses, and e-commerce platforms that need product models. The generated assets are built to slot into the standard tools of the trade: Unity and Unreal Engine, via plugins and an API that Kaedim launched, characteristically, through Y Combinator before its big raise.

The money arrived in March 2024: a $15 million Series A led by a16z Games, the games-focused arm of Andreessen Horowitz. What's notable is less the number than the cap table. Alongside the fund, the round drew a roster of operators who tend to only write checks when they believe the underlying thesis - Nate Mitchell, a co-founder of Oculus; Tom Blomfield, who founded Monzo; Patrick Pichette, the former CFO of Google; Scott Gelb, formerly president of games at Riot; and several others. That's a lot of people who have built large things choosing to bet on 3D content becoming a bottleneck worth removing.

Where it sits, and what could go wrong

Kaedim is not alone in this race. A wave of text-to-3D and image-to-3D tools - Meshy, Tripo, Luma, Rodin, and others - are chasing the fully automated version of the same dream, and the underlying models are improving quickly. Kaedim's bet is that "fully automated" and "production-ready" are, for now, two different things, and that the artists in the loop are a feature rather than a cost to be engineered away. If the models get good enough fast enough, that bet gets harder. If they don't - if the last 10% of quality stays stubbornly human for a while - Kaedim's approach looks less like a workaround and more like the correct architecture.

What's clear is that the company has moved past the phase where its story was mostly about a controversy. Four years in, Psoma marked the anniversary with a note about building the future of 3D content, and the framing had settled: not a magic box, but a pipeline. For a category that spent 2023 and 2024 promising more automation than it could deliver, there's something quietly useful about a company whose hardest lesson taught it to describe exactly what it does. The 3D content problem is real, the budgets prove it, and Kaedim's answer - machines and people, in a loop - is at least an honest one.