The newly released free Intel® Game Dev AI Toolkit for game developers combines the power of AI inferencing with add-ins for style transfer, object detection, and world creation, all available as a standard Unity* Editor package. The release represents a significant step forward as Intel continues to provide game developers with leading-edge tools that harness the power of Intel® architecture.
Figure 1. The first version of the AI Game Dev Toolkit
The toolkit consists of foundational classes that asset creators and game developers can build on to optimize, tune, and run comprehensive AI inference workloads across a broad range of platforms and architectures. The goal is to harness machine learning and AI-enabled capabilities to elevate game development across the ecosystem, ultimately delivering great gaming experiences to the hundreds of millions of PC gamers around the world.
AI Inferencing and Rapid Creation of Game Worlds
One of the key ingredients in the inaugural version of the toolkit is Gaia ML, the machine-learning version of Gaia produced by Australia-based company Procedural Worlds. Gaia ML enables the creation of richly textured game worlds in mere minutes—worlds that feature complex and photo-realistic terrains, lush vegetation, and lifelike environmental systems.
Intel vice president Roger Chandler previewed the Intel Game Dev AI Toolkit in late 2021 and was excited about the inclusion of Gaia ML, writing that it promises to elevate game-world creation capabilities to a scale previously unimaginable. “We’re excited to partner with Procedural Worlds, which allows us to combine the power of AI inferencing with their well-established (and developer-friendly) Gaia interface,” he said.
Adam Goodrich, founder and general manager at Procedural Worlds, was equally enthusiastic about incorporating Gaia ML into the Intel Game Dev AI Toolkit. "Procedural Worlds is excited to collaborate with Intel to bring the power of machine learning to the game-development community,” Adam said. “Gaia ML makes world creation easy and fast, and by bringing this same ease of use to the creator community, we lower the barrier of entry to people looking to do interesting things with AI and machine learning (ML) in their game worlds,” he added.
What's Included
The Intel Game Dev AI Toolkit contains the Intel® Distribution of OpenVINO™ toolkit, an open-source package for optimizing and deploying AI inference. It can boost deep-learning performance in computer vision, automatic speech recognition, natural language processing (NLP), and other common tasks. Intel Distribution of OpenVINO toolkit helps target the right accelerators on the CPU or GPU—or both simultaneously—and works with popular frameworks, such as TensorFlow* and PyTorch*, to reduce resource demands on a range of Intel® platforms, from edge to cloud.
By using Intel Distribution of OpenVINO toolkit, game developers can implement GAN-based style transfer to stylize entire scenes. Alternatively, game developers can focus on specific objects and characters in a scene for extra artistic flexibility. The toolkit also unlocks the potential of in-game object detection for developers, opening up possibilities such as AI-assisted target identification. Both the style transfer and object detection components of the toolkit are based on research championed by Intel® Software Innovator Christian Mills, and code that he optimized to run in the Unity engine.
The Intel Game Dev AI Toolkit removes the need for developers to understand internal inferencing mechanics to accelerate machine learning, which is performed by the Intel Distribution of OpenVINO toolkit. The toolkit relies on the inference engine (whether it’s OpenVINO, Unity* Barracuda, or another) to do its magic. Without the toolkit, developers would have to hand-code features, which could potentially take months.
In addition to resources available at launch, including videos, sample code, and demo projects, the Intel launch team will continue developing new training material as needs materialize.
Training Resources for Best Success
Both Intel and Procedural Worlds will offer training resources to help new users get started. In addition to resources available at launch, they include videos, sample code, and demo projects. Procedural Worlds also offers training videos and tutorials for creating game worlds on their YouTube* channel. As the top-rated world-, landscape-, and scene-creation tool in the Unity asset store, they have over 100,000 customers worldwide, using Gaia, Gaia Pro, GeNa Pro and now Gaia ML.
Figure 2. The Gaia ML interface offers users a powerful tool for creating beautiful game worlds.
Long committed to sustained growth among indies, studios and professionals, Procedural Worlds recently launched Canopy, a “powerhouse of resources” that works on a subscription basis, with tools, tutorials, and assets all in one place. Adam Goodrich predicts that Intel Game Dev AI Toolkit users could easily learn to generate a world with Gaia ML, and then follow a setup wizard to add their game into that world. Next would come the addition of machine learning, so the user could demonstrate typical object detection and style transfer examples in a complete game.
Intel and Procedural Worlds to Co-Present At Game Developers Conference (GDC) 2022
To spread the word about the new toolkit, Intel and Procedural Worlds will share the stage at a GDC 2022 session showcasing the project. Intel’s Peter Cross will also join Adam Goodrich for a 30-minute presentation at the Unity Theater at GDC.
Adam Goodrich's presentation idea involves rapid construction of a 3D tower-defense game with a rail shooter, built from mix-and-match templates available via a Canopy subscription. Adam could add in machine learning-based object detection, and then have the game adapt itself to what it is seeing—and it would be open to the global student community. Adam hopes to make such templates available through Canopy soon.
Intel’s goal is to make the Intel Game Dev AI Toolkit accessible to any non-AI or machine-learning experts who are ready to get started. Newcomers can use version 1.0 as a way to familiarize themselves with machine-learning concepts and take them to another level. New machine learning add-ins and capabilities should arrive over time, and version 2.0 of the toolkit is tentatively slated for late 2022.
Download the toolkit to begin incorporating AI and machine learning into your projects. If you haven’t yet begun to incorporate AI and machine-learning technology into your projects, now is the perfect time to start.
Additional Resources
In-Game Style Transfer Tutorial Part 1
AI Inferencing and Object Detection Tutorial
Procedural Worlds Canopy “You Ask–We Make” Offer
Procedural Worlds “Work in Progress” Report from the xChange in Canopy
Rails-based Aerial Shooting Game Template from Procedural Worlds