Roblox AI tool generates Clair Obscur lookalike, sparking plagiarism concerns

Roblox AI Generates Controversial Game Copy

The Roblox Corporation announced its experimental generative AI technology on February 5, unveiling what it calls a "real-time, action conditioned world model" designed to help users create playable game worlds from simple text prompts and image prompts1

1

. The demonstration, however, immediately drew criticism when the company showcased footage that appeared to directly copy Sandfall Interactive's critically acclaimed Clair Obscur: Expedition 33, winner of 2025's Game of the Year2

2

.

The reveal came as an unexpected addendum to the Roblox Corporation's primary announcement about 4D Generation, a feature for the Cube Foundation Model that allows creators to generate interactive 3D objects like cars and planes through prompts1

1

. While 4D Generation launched yesterday for use in Roblox Studio, the real-time world model remains in the research phase at the company's lab2

2

.

Real-Time Dreaming Raises Plagiarism Questions

The controversial demonstration showed what Roblox calls "real-time dreaming," where a prompt as simple as "woman in a glowing cave" generated what users immediately recognized as Expedition 33's Flying Waters location, complete with the character Maelle's distinctive model and outfit1

1

. The company envisions this technology enabling what it calls "Dream Theater," where one user generates worlds while others watch and provide additional prompts2

2

.

User concerns about potential copyright infringement surfaced immediately, with one viral response asking Sandfall Interactive directly: "What do you think about this? I wonder what you trained your model on?"1

1

. The resemblance appears too specific to be coincidental, raising serious questions about the sources used in AI model training data.

Training Data Sources Under Scrutiny

Roblox co-founder and CEO David Baszucki explained that the company "built this model with internal Roblox data, as well as open source video data"2

2

. The official Roblox X account clarified that the world models have been "trained on a combination of data, including proprietary Roblox 3D avatar/world interaction data"1

1

. That internal training data includes information from the 13 billion hours Roblox players spend in-game every month, which is now being used to train world models and AI-driven NPCs2

2

.

The word "including" in Roblox's explanation appears to acknowledge that other data sources were used, though the company has not disclosed what open-source video data went into training the model1

1

. This lack of transparency mirrors broader industry concerns about how generative AI companies source their training data, particularly when the output so clearly resembles copyrighted material.

Industry Implications and Future Concerns

The incident highlights the ongoing tension between rapid AI development and copyright protection in the gaming industry. Roblox's experimental generative AI technology functions similarly to Google's Project Genie, which launched last week and faced criticism for generating content resembling Mario and Zelda1

1

. The company envisions using these world models in Roblox Studio, where creators could "walk around and use prompts to 'paint' a world and then convert it into a 3D representation" that multiple people can play simultaneously1

1

.

As Roblox pushes forward with plans to integrate this technology into its platform used by billions, the industry will be watching how developers like Sandfall Interactive respond to what many perceive as plagiarism. The quality of the AI-generated Clair Obscur lookalike also drew mockery, with users noting the blurry, disorienting visuals that one commenter described as making the Game of the Year winner "look so bad"2

2

. Whether this represents the future of user-generated content or a legal minefield remains to be seen.