newsletter Newsletter

Promethean helps video game designers with an AI assistant that automatically assembles scenes in the designer's style. The startup is now introducing new features and a virtual reality version.


In 2018, Andrew Maximov, former art director at Naughty Dog, developed the idea for Promethean AI, an AI system for designing virtual worlds. Promethean AI is named after the titan Prometheus - creator and protector of mankind.

For Maximov, AI is a crucial step in keeping the ever-increasing cost of video game production in check. In 2017, he predicted production costs of up to $200 million for triple-AAA titles. Studios would then need to sell seven million copies to be profitable. This gigantic financial hurdle is a catalyst for the exploitation of workers and questionable game design, such as loot boxes.

Promethean AI is designed to save video game designers time by automating repetitive tasks and aiding design with intelligent suggestions and quick prototypes. The team first unveiled its product to the public in November 2018. Now, Promethean is providing an update on the software's current state and a preview of new features.


Automatic asset remixer

In the Promethean AI keynote, the team showed numerous examples of how the AI system and Unreal Engine can be used to create high-quality visual environments in minutes, including scenes such as an idyllic wooden house on the edge of a forest, a modern apartment, and a gas station.

The examples provide a glimpse into Promethean's workflow: the designer accesses assets from a central console, displays design alternatives, or generates entire parts of a scene and reworks them rather than developing them from scratch.

For Promethean's AI to work, designers must import their asset library into the software. The AI then automatically assigns tags and sorts objects. Promethean uses existing metadata and image recognition algorithms to do this. All information can be manually customized.

Once assets have been imported, they can be accessed and placed via the console. Promethean can also select objects in a scene, suggest alternative assets or materials, and respond to requests such as "Pick something to drink" or "Show me some fruit. This means that artists no longer have to interrupt their creative process by searching through object libraries, the startup says.

The ability to recognize assets and offer alternatives is included in the free version of Promethean AI, which has been available for a few days. According to Maximov, it is the foundation of Promethean's more impressive features.


AI world generator

The more advanced functions of the AI system are currently reserved for Promethean's paying customers. Maximov is talking about levels 3 and 4: the AI pre-filters objects according to context and can independently position them in compositions.

For example, if the level designer is building a kitchen in a certain style, Promethean will recommend appropriate assets such as refrigerators, dishes, or food. In a garage, the AI filters for metal cabinets, workbenches, cars, or tools. This contextual filtering also works for individual objects: If the artist selects a desk, the AI suggests computers, typewriters, magazines, photos, or a telephone.

Promethean can also classify objects by time: A typewriter is older than a notebook, and a smartphone is newer than a dial phone. This allows artists to sort their proposals according to the desired era and create themed environments even faster. The free version of Promethean will get these level 3 features later this year.

Level 4 goes a step further: Promethean can independently assemble detailed compositions upon request. A desk is equipped with a computer, a picture frame, magazines, a telephone and a paperweight. A workbench is fully equipped with neatly arranged tools.

Join our community
Join the DECODER community on Discord, Reddit or Twitter - we can't wait to meet you.

Promethean learns such compositions from the artists who use the software. In one demo, the team shows how a campground created by an artist is saved as a template. Rather than simply copying the exact layout, Promethean learns the semantic structure of a scene: there are tents with certain objects, such as water bottles and backpacks, placed inside. There are chairs in front of the tent, and a fire is burning. The details of this semantic structure can be customized by the user.

Promethean can then generate a campsite that adheres to the predefined structure but looks different every time. In the demo, this works quite well, but here and there there are unfortunate asset placements, such as a chair standing inside another chair, or a backpack shimmering through a tent wall. Such mistakes can be quickly corrected by the level designer.

The next logical step is the complete generation of digital worlds. The startup describes this development goal as stage 5, which would turn Promethean into a transport vehicle that could take virtual visitors to any imaginable world - provided they have the appropriate training. But Promethean is not quite there yet.

Asset browser, collaboration tool & VR application

In addition to the AI assistant, Promethean offers a comprehensive asset browser that allows users to sort all imported assets by a variety of categories and display formats to find the perfect asset for the job.

For team collaboration, Promethean includes the Pallet, a platform where artists can upload and download groups of assets via the cloud, share information and store design templates in image or video format.

At the end of the keynote, Maximov showed a prototype of Promethean's AI for virtual reality. Here, an artist uses voice commands to set up an office, changing the view from the window, the era, the wallpaper, the computer, and the posters on the wall.


The interface is still a bit clunky apart from the voice input, but otherwise, the level design via voice input works surprisingly well thanks to Promethean's understanding of context. Maximov does not reveal if and when Promethean plans to release a VR version.

Promethean AI currently supports Unreal Engine 4, with support for Maya, Blender, and 3ds Max coming soon.

Support our independent, free-access reporting. Any contribution helps and secures our future. Support now:
Bank transfer
Max is managing editor at THE DECODER. As a trained philosopher, he deals with consciousness, AI, and the question of whether machines can really think or just pretend to.
Join our community
Join the DECODER community on Discord, Reddit or Twitter - we can't wait to meet you.