Traditional game development requires manual asset creation, coding and testing, taking at least 3-12 months. With generative AI (GenAI) you can build games with simple text prompts, slashing development time. How soon will GenAI allow games to be built on the fly?
What’s new in AI-assisted games?
Over the last couple of years, creators have begun using GenAI tools like GPT, DALL-E, Stable Diffusion, Unity’s Muse and Nvidia’s ACE to create scenes, characters, code, dialogues, music, art and gameplay mechanics with just natural language prompts. For instance, Ubisoft (Assassin’s Creed, Prince of Persia) uses these tools to generate in-game dialogue, while startups Scenario and Leonardo.Ai help individuals and small developers (collectively known as ‘indie’) create game-ready textures, characters and environments with prompts. Game engines like Unity and Unreal Engine then automate these elements to build a game.
Also read | GenAI to boost India’s IT industry’s productivity by up to 45%, EY India survey shows
Are there any big credible examples?
While fully AI-developed large-scale games don’t exist yet, Oasis by Israeli AI startup Decart showcased real-time AI-generated frames when released last October. Nvidia had earlier recreated the classic game Pac-Man using AI without a game engine. Last October, Oasis demonstrated an entirely AI-generated game with all actions such as moving, jumping, collecting items and breaking blocks generated by AI in real-time at 20 frames per second (fps), again without a standard game engine. This month, Roblox announced Cube 3D—an open-source AI tool that creates 3D objects and scenes from text.
What does AI automation portend for developers?
AI can create game assets, scripts and environments in minutes, reducing development time by 30-50%. While automation may shrink junior roles, the demand for AI specialists, designers and narrative developers will rise, with a greater focus on supervising AI, refining mechanics and storytelling. And anyone can build and monetize a decent game.
Also read | AI power shift: Can China close the gap with the US?
What are the current limitations?
AAA-quality real-time games need better AI reasoning, consistency (to avoid a disjointed experience) and real-time adaptability. Further, AI-built Oasis runs at 20 FPS which is patchy, while top games require 60-280 FPS for smooth running. Open-world games like Dragon’s Dogma 2 and multiplayer titles like Counter-Strike 2 demand extensive testing and optimization. AI-generated content also needs human oversight for quality, coherence and emotional impact. A lot of computing power is also needed.
Will AI ever create a good game by itself?
AI can already create gaming assets from scratch without a game engine by training on vast datasets. Roblox CEO David Baszucki says the company is developing a 3D foundational model trained on thousands of in-game assets like castles, vehicles and creatures. This will enable AI to generate detailed environments and characters quickly. Roblox’s physics simulations enhance realism. And despite the limitations cited above, Nvidia CEO Jensen Huang predicts fully AI-generated games in 5-10 years.
Also read | Adobe seeks to sell lawsuit-free AI models, expands Indian engineering hub
