Last week, AI startup Luma posted a series of videos created using its new video generation tool, Dream Machine, which the company describes as “a highly scalable and efficient Transformer model trained directly on video.”
The only problem is that at the 57 second mark, the Dream Machine generates Monster Camp Pixar's Mike Wazowski features a slightly blurry AI, but still recognizable character. Monsters Inc Many noticed that multiple characters and the overall aesthetic look were borrowed from the franchise, and questions immediately began pouring in.
Was it given prompts for Pixar-style animation? Was it trained on material that includes work from Disney Studios? This lack of transparency is one of the biggest concerns about these kinds of models, and Dream Machine is one of a number of text-to-video AI tools unveiled in recent months, following OpenAI's Sora and Google's VideoPoet and Veo.
Luma touted its Dream Machine model as the future of filmmaking, featuring “high-quality, realistic shots” that could be created by simply filling in prompts in a box. Between videos of cars on a melting highway and awkwardly narrated sci-fi shorts, it's easy to see why tech enthusiasts were quick to call it a radical innovation.
Currently, Luma invites users to sign up and play Dream Machine for free, but the company also has “Pro” and other levels that will charge users for more features. Luma did not respond to a request for comment on the source of the footage it uses for training on Dream Machine at the time of publication.
Disney hasn't commented publicly on what Luma is up to, and it's possible the company isn't even aware of it, but at a time when there are calls for greater transparency about the datasets that power AI tools like the one Luma is building. Monster Camp It’s hard not to see the generative AI ecosystem as prone to plagiarism.
Correction, June 18: This article originally incorrectly stated when the AI video was first posted; it was posted last week, not over the weekend.
