• James Kinch

Modl.ai reveals automatic Match-3 Puzzle maker

Modl.ai, a development group focused on using AI tools to assist in game creation, has revealed the "Puzzle Maker". A tool that creates match-3 puzzle levels using artificial intelligence.

Whereas a specialist level designer might be able to design one or two levels in a day, Puzzle Maker can generate hundreds of levels in just a few days, hugely reducing the human time and cost of supporting these games.

Match-3 games are some of the world’s most popular games - particularly on mobile - with examples like Candy Crush Saga, Gardenscapes and Bejeweled Blitz downloaded hundreds of millions of times. According to game analytics company GameRefinery, match-3 games are so popular that they make up 16% of all mobile game revenues on the US Apple app store.

To give some idea of how important level creation is to these games, Candy Crush Saga currently has more than 9000 levels - with more added on a weekly basis. Any ability to automate the level design and creation process therefore has the potential to save developers considerable time and money.

Unlike other types of puzzle, match-3 games don’t have a single solution - meaning they are very hard to create using automated software. modl.ai’s approach was to create two AI-based tools: a content generator and a level tester that, when used together, solve many of the challenges of creating a constant flow of game levels.

Puzzle Maker’s content generator is based on procedural content generation via machine learning (PCGML), meaning that it learns from existing levels and also improves the more levels it creates, with as few as 100 levels needed to complete the training phase. Each level the tool creates can be customised through a dashboard where the level designer can set a variety of parameters, from the size of the playing canvas to the kinds of puzzle pieces and challenges in each level.

Once levels have been generated, Puzzle Maker tests each level with AI bots to understand how many moves it takes to complete a level, the time it takes to complete, the number of reshuffles required and even the predicted likelihood that players will win the level - a key part of keeping players engaged over time. This data provides developers with insights into each level’s difficulty and provides characterisations of the structural level features such as the ratios of piece types and their distribution. All of this information is displayed to the level designer, who can review each one and decide if they are good enough to be included in the actual game. If they are, the code for the level is generated ready to be added.

“Match-3 games are often seen as simple and only for casual gamers, but in reality they are very clever and complex. This means that developers have relied on skilled level designers to keep coming up with fresh designs, and the only way to make more levels is to hire more level designers. We saw an opportunity to bring a new approach,” said Christoffer Holmgård, CEO at modl.ai. “By giving level designers a tool like Puzzle Maker, they get to use their expertise to know what their audience will love, but without the hours of trial and error work that we’ve automated. By solving the quantity problem, developers can focus on the quality.”