Assaf Vaknin.

"Israeli companies are responsible for 5% of global mobile game downloads"

Assaf Vaknin, Head of Apps at Google, spoke at Calcalist's Gaming Conference in partnership with Google and Playtika about how the world of gaming, while Dani Valevski, a researcher at Google Research discussed the future of gaming

“The year 2024 is expected to end with a 6% decline in mobile gaming user spending globally, primarily due to a drop in disposable income. When diving deeper and comparing Israel to the rest of the world, the decline is even greater,” said Assaf Vaknin, Head of Apps at Google, at Calcalist’s Gaming Conference, held in collaboration with Google and Playtika. However, Vaknin noted that the decline is not particularly concerning, as it mainly stems from the hyper-casual gaming sector. According to him, the key takeaway is that “Israel still accounts for 5% of global downloads.”
“When examining the sources of this substantial decline, we see that the most significant drop is in hyper-casual games, as well as in racing and simulation games. People download games to reduce stress, so the decline in these categories in Israel is understandable - we’ve had enough action this year. On the other hand, there’s an increase in puzzles, word games, casino games, card, and board games.”
2 View gallery
כנס גיימינג - אסף וקנין ראש תחום אפליקציות Google וידאו
כנס גיימינג - אסף וקנין ראש תחום אפליקציות Google וידאו
Assaf Vaknin.
(Credit: Yariv Katz)
He noted that despite major successes in the sector this past year including Playtika’s acquisition of SuperPlay, Plarium’s acquisition, CrazyLabs reaching 7 billion downloads, and VC vgames raising $142 million last week, challenges remain.
What needs to be done to level up? Vaknin identified three main areas: expanding into Asia; hybrid monetization; and leveraging celebrities and building brands. He noted that Israeli companies are highly focused on the U.S. and Western Europe, leaving untapped users and dollars in Asia. “Companies that successfully break into major markets like Japan and South Korea, such as Roblox, significantly boost their revenues,” said Vaknin.
In regards to hybrid monetization, he says that there needs to be a combination of advertising revenue and in-app purchases. “I’m happy to see many Israeli companies beginning to move toward this model. It’s possible to achieve meteoric revenue growth with a tailored game economy.” Finally, Vaknin says that the impact of collaborations with celebrities “lies in creating a stronger connection with users. Israeli creativity has room to take this concept further to increase revenues.”
The future of gaming
Also speaking on the subject was Dani Valevski, a researcher at Google Research, who discussed a paper he co-authored on how to transform an image-generation model into an engine for video games. “Doom is an iconic game from 1993 that made the genre popular. The plot is very simple: a soldier fights demons who have taken over a base on Mars. It featured detailed and rich levels, doors that could be opened, stairs, monsters with different abilities, and an interactive environment where you could shoot barrels or be harmed by acid. What made it possible was a game engine that became a milestone in programming and gaming. A game engine refers to the code that runs the game. The engine processes maps and graphics, manages game logic, and displays everything in 3D,” he explained.
2 View gallery
כנס גיימינג - דני ולבסקי חוקר Google Research וידאו
כנס גיימינג - דני ולבסקי חוקר Google Research וידאו
Dani Valevski.
(Credit: Yariv Katz)
“We asked ourselves: what if we ran a game engine entirely on a neural network, end-to-end? The idea is that everything the complex engine knows how to do would be managed by the model. The goal is for the model to generate this automatically by watching recordings of the game. Every frame is generated by the model. We sent the video to John Carmack, the creator of Doom’s game engine.”
How do you teach a model to simulate Doom without writing a single line of code? Valevski explained that they adapted a text-to-image model to accept keyboard inputs instead of text and produce the corresponding game frame. “If we can run this in a very fast loop - player keyboard inputs leading to frame generation - we can teach the model to predict frames automatically so that what’s shown on screen matches the player’s actions.”
“We had to feed the model many hours of gameplay, and the solution was to use a bot to play the game. Automated gameplay has been researched in the past,” he explained. They used a bot designed to explore the game as much as possible and reach as many states as it could. “We optimize the model without writing a single line of code. We published the paper three months ago, and it went viral quickly. We showed that an image-generation model can run a video game for extended periods without quality degradation.”
“In essence, we transferred an existing game to run on a neural network. The exciting part is that we might use neural networks to generate new behaviors or even entirely new games,” he said.
According to Valevski, the implications and future possibilities of this research are endless. “What if we applied this to thousands of games? Could we combine games? Another avenue is using video scenes from new games to generate new behaviors based on images alone. In the future, we might even create entirely new levels this way. There’s an opportunity to develop capabilities that allow us to generate games ourselves.”