Unity
Muse Chat
Unity’s flagship AI product, Muse Chat, is the crown jewel of Unity’s machine learning sub-brand, Muse. On an extremely accelerated timeline our small, multidisciplinary innovation team productized a Unity-centric GPT, progressing from ideation to launch in mere months.
The 0-to-1 productization for Muse was an incredibly collaborative and intensely expedited effort. Under Charles Migos (VP, Design), I orchestrated our talented Design team’s interface development—incorporating Muse branding elements from Marc Cinq-Mars (VP, Brand), and team nearly as soon as they were conceived. Our regular cadence with senior executive leadership, including Sylvio Drouin (SVP, Innovation), Luc Barthelet (CTO), Carol Carpenter (CMO), and Marc Whitten (President, Create) ensured that we remained aligned despite Muse Chat’s breakneck pace.
Muse Chat would literally not have been possible without our friends in Engineering, Product, Marketing, and Biz Ops. In particular we owe a debt of gratitude to Nico Perony (Director of AI Research), and Martina Johannesson (Senior Technical Product Manager). Last, but most dear to us on the Design side (and our secret weapon) Liz Pomeroy (Senior Product Designer) who not only juggled interface bits while working well outside her time zone, but often lead the executive conversations with a confidence that assured this absolute toughest crowd.
Code generation
Unity Muse supercharges your project development. Use it to generate new C# sample code, check snippets for errors, and troubleshoot as you go. Muse can even explain the concepts and intent behind the code it generate, as in this example. Here we’ve asked Muse, “Can you show me how to set up a character controller that can walk, run, and jump?” Muse responds with version-specific instructions for adding a new component, configuring it with relevant parameters, and even provides working source code to attach to the component.
Muse ecosystem
As we mature our Muse product line, each Muse ability (from Chat, to Texture, to Animate, and beyond) will become increasingly aware of its sibling abilities. Imagine launching a pre-populated Muse Animate instance from a prompt written in Muse Chat, or bouncing between Muse Animate and Muse Behavior as they inform each other of your evolving intended interactivity. Muse Sketch, previewed at Unite 2023, only hints at where Muse is headed. And you can bet that all of these Muse abilities will grow more deeply imbedded into your everyday Unity Editor workflows.
Charles Migos and Liz Couto preview Unity’s Muse Sketch during the Unite 2023 keynote.
Responsible model training
As we venture further in developing our own machine learning models, we’re taking the ethics of model training very seriously. For example, in order to provide useful outputs that are safe, responsible, and respectful of other creators’ copyright, we’ve pioneered two bespoke diffusion models—one for Muse Texture and one for Muse Sprite—each trained from scratch on proprietary data that is Unity-owned or licensed. We also know that having a consistent look and feel for your game is essential. Muse Sprite’s Style Training makes this easy by ensuring your game has a unified look throughout.
Responsible and consistent; this is only the beginning on the conversation. We’ll have a lot more to say about Muse Chat model training, as well as other products and models, in the near future. For additional editorial background see “Responsible AI and enhanced model training at Unity” by Sylvio Drouin (SVP, Innovation) and “Updating Unity’s guiding principles for ethical AI” by Marc Whitten, (Chief Product and Technology Officer). For common questions and answers related to datasets, see “Unity AI: Frequently Asked Questions.”