EXCLUSIVE: Disney and NBCUniversal accused Midjourney of being a “bottomless pit of plagiarism” in a lawsuit last year. More than 18 months on, the company has allegedly failed to clean up its act. VN.ai, a tech startup that stress tests generative artificial intelligence tools for copyright infringement, has produced its first quarterly report examining how 15 popular AI models performed when it comes to flouting studio IP. The research conducted last month found that Midjourney v8.1 was the worst offender when it came to reproducing images of copyrighted characters. The AI model also outranked competitors on so-called “brand safety violations,” meaning characters were placed in a “harmful context,” such as scenes of violence or drug taking. According to VN.ai’s report, which was shared with Deadline, Midjourney reproduced images of characters that violated brand safety measures once in every three text prompts, or 37% of the time. This included a Midjourney v8.1 hallucination in which it generated an image of The Joker smoking cannabis while wearing a dress that could have been ripped from the wardrobe of Frozen’s Elsa. Watch on Deadline Midjourney did not respond to requests for comment. Disney and NBCUniversal’s lawsuit against the company remains ongoing. The startup has argued that training its AI models on images of copyrighted material is permitted under fair use. VN.ai’s report found that every AI model generated at least one copyright-protected character, even when a character’s name was not referenced in a text prompt. This included Google’s Nano Banana and ByteDance’s model Seedance 2.5, though the latter has significantly improved after receiving cease-and-desist letters from Hollywood studios. As noted recently by Deadline, other Chinese AI models continue to plunder IP, including MiniMax H3, which generated videos that violated studio brand safety standards 55% of the time when an image was used in a prompt. Asked why AI models continue to flout copyright, VN.ai founder Nelson Chu said it was “good for business.” He added: “It’s great marketing. It works. It gets people to use your model and pay for it.” Chu predicts change, however. “I do think at some point that’s just not going to be okay. And I think as more models set the precedent for how it is to be a good actor here, you will see more improvement,” he told Deadline.
Midjourney: Report Claims AI Model Is Worst Offender For Flouting Hollywood Copyright
Full Article
Original Source
Read the full article at Deadline →KhanList aggregates and links to publicly available news content. We do not host full articles from third-party sources. Always verify important information with original sources.