Movie2Game-Bench tests how AI models adapt films into playable games
A new arXiv benchmark evaluates film-to-game adaptation beyond playability, scoring 14 models across 22 films and finding shared weaknesses in playable sequences.
In plain words
Researchers released a benchmark that grades how well AI systems turn a movie into a working game, checking not just whether the game runs but whether it keeps the film's content. It uses 22 films, hundreds of reference items, and automated players that test the games. Across 14 models, scenes built for actual play scored lowest.
What was announced
The benchmark, called Movie2Game-Bench (M2G-bench), is described in a new arXiv paper. Its authors say adapting a film into a game means choosing source elements and translating them into player actions, rules, and feedback. The evaluation protocol combines general assessment, movie-specific assessment, and playtesting by three GUI model players, and it links source content to gameplay steps and implementation evidence so both content preservation and interactive realization are captured. The benchmark covers 22 films with 658 source-specific reference items.
Fourteen models were evaluated on the same films using their selected primary harnesses. According to the paper, the results show complementary strengths in game quality, adaptation, and player experience, but a shared weakness in playable sequences: that category had the lowest content-presence mean in all 14 models, averaging 37.9 out of 100 against 54.8 for iconic elements. Among judgments of retained emotional-development content, narrative choices made up 60.0% and scenario operation 7.2%. The authors point to two linked priorities: preserving source situations that can support gameplay, and connecting retained content to player actions and consequences.
What it does not show
- The source is the paper's abstract only, so evaluation methods, model identities, and full results are not available for verification.
- The reported scores and percentages come from the authors' own benchmark and have not been independently confirmed.
Sources
Source-support review completed 20 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.
Coverage: arXiv: abstract.
Updates and corrections
No corrections so far.
First published 20 Nov 2026
New to the subject? AI from the Ground Up starts from the beginning.