Games are traditionally crafted by hand by a group of people with very different expertise. Most of the time goes into content creation. How will generative AI change the balance of effort in game creation?
We explore existing types of game creation. We focus on procedural content generation, which uses a small random seed to create seemingly indefinite volumes of content. A new AI model called Genie addresses the same problem using deep learning and computer vision. Genie can generate entire games from scratch.
However, in the short term it's more likely that AI models will assist game developers in their creation process. One example of this is a terrain generation paper that appeared at AAAI 2024. It is interesting to imagine what creations will be possible in the future with a lower barrier to entry.
#ai #games #contentgeneration
Google DeepMind’s new generative model makes Super Mario–like games from scratch
[ Ссылка ]
Genie: Generative Interactive Environments [paper]
[ Ссылка ]
MindBranches X Post on Genie
[ Ссылка ]
Terrain Diffusion Network: Climatic-Aware Terrain Generation with Geological Sketch Guidance [paper]
[ Ссылка ]
.kkrieger - The FPS Game in 96KB (Full Gameplay)
[ Ссылка ]
Infinite craft [game]
[ Ссылка ]
Dungeon crawl stone soup [game]
[ Ссылка ]
NetHack [game]
[ Ссылка ]
0:00 Intro
0:22 Contents
0:30 Part 1: Anatomy of a game
0:41 Open world vs scripted games
1:04 Generative AI can tailor whole universe
1:43 Example: dynamic AI engine in Left4Dead 2
2:06 Addictive potential of generative AI games
2:51 Games are technically challenging to create
3:14 Game engine description
3:35 Huge number of other skills needed
3:47 Not a lucrative industry unfortunately
4:20 Part 2: History of content generation
4:36 Most effort goes towards content, not game engine
5:20 Hand crafted content vs procedural generation
5:51 Random seeds for procedural content generation
6:25 Traditional assets take a lot of space
6:47 Example: .kkrieger game less than 100KB
7:26 Example: Elite Dangerous populates galaxy
7:48 Example: No Man's Sky generates flora, fauna, etc
8:12 Example: Minecraft needs only one seed
8:30 Example: nethack and crawl, highly inventive
8:40 Using algorithms (genetic, etc) to generate levels at dev time
9:42 Example: interactive terrain generation
10:45 Part 3: Generative whole games
10:53 Genie paper from Google
11:11 Platformer game dataset, 7 million samples
12:09 Predicting future video frames (computer vision)
12:45 Enough similarities between platformer games
13:24 Trying to make the research reproducible
13:55 Asking for scenarios in English text
14:15 Model can even emulate parallax
14:59 Are all games about to be fully algorithmically generated?
15:22 Game generation might enable more inventiveness
15:59 The game itself might just be generative AI
16:21 Example: infinite craft
17:03 Conclusion
17:25 Procedural generation
18:14 Generating entire games from scratch
18:57 Video of Genie
19:28 Outro
