Newborn Town's TopTop Launches AI Mini-Game Built on Viggle's Physics-Native World Model

A player in Japan sat down to a mini-game called Slap Showdown and, over 22 days, finished 1,041 matches. On his best day he played 151. Averaged out, that is a match every 9.5 minutes. He is the most vivid data point in the account TopTop has given of its newest game, and it comes from the company's own operations team.

TopTop is the game-oriented social platform run by Newborn Town. Slap Showdown, launched in September, is an AI-powered 3D mini-game built with Viggle, a company that specializes in physics-native world models. The game runs on Viggle's self-developed Joint Space-Time model, known as JST. Players upload a photo from their phone and, in around 30 seconds, get a customized 3D character: an anime protagonist, a cyberpunk cyborg, a LEGO minifigure. They then take that character into multiplayer battles against other users.

The controls are press-and-hold, tap and drag, and the characters answer with physics-based movement. The point, as TopTop describes it, is that the AI-generated figure is part of the gameplay and not only a cosmetic skin. The game has gained particular traction in Japan, among Spanish-speaking users in North America, and across the Middle East, and it is now available in multiple countries and regions.

The numbers on offer come from TopTop's operations team. In the game's first month, it says, Slap Showdown drew a growing base of highly engaged players with high spending potential. It reports a 97% match completion rate and a game-start rate above the platform-wide average. Beyond the Japanese player, it cites an avid player in Eastern Europe who completed 735 matches in 11 days, averaging 66.8 a day with a peak of 126 in one day. None of these figures has been independently verified.

The strategy behind the game is stated plainly. On a platform driven by user-generated content, mini-games act as entry points and traffic drivers, while social features carry retention and monetization. Personalized 3D avatars, TopTop argues, give users new ways to express themselves and share moments with friends. Its operations team says future releases could extend AI 3D generation from avatars to customized environments, and that existing games could be refreshed with new gameplay features. It also suggests world models could eventually enable more immersive, face-to-face interaction in virtual spaces.

Viggle was founded in 2022 and builds physics-native world foundation models. It first drew wide attention with Meme Maker, a tool that animates static characters and generates videos from specified movements. In August 2024 it raised USD 19 million in a Series A led by Andreessen Horowitz. On 19 October, it fully launched JST-2, which it compares in significance to GPT-2's place in the history of language models.

Where many world models infer physical structure indirectly from pixels, JST-2 is built to learn the structure and dynamics of the 3D world directly. Its underlying representation uses units resembling digital atoms, each with a position in three-dimensional space, physical properties such as material, color and mass, and information on how it moves and responds to force. After four years of research, Viggle says one architecture can generate characters, motion and scenes. It claims a 3D scene can be produced in seconds at an inference cost under one US cent, and that finished content can be rendered locally without rerunning the model for each frame.

Viggle's next model, JST-3, is meant to be a comprehensive world simulator that could underpin game and graphics engines and serve as a virtual training ground for robotics and other physical AI. The longer-term aim is for it to learn physical rules from data rather than from rules written by developers. Newborn Town, which has invested in Viggle, also backs projects in AI companionship, agents and interactive gaming. The open question is whether Slap Showdown's early engagement holds up and can be confirmed beyond the company's own reporting.