"Back when generative AI first reared its ugly head, there was a panic that the average viewer wouldn't be able to identify it. But now, years into the rise of OpenAI, Grok, Midjourney, and the myriad other AI agents, anyone who has spent time online can identify a specific "AI look": weirdly flat lighting, nonsense text, and often a strange yellowish tinge."
"In March 2025, it was all the rage to "Ghiblify" your photos using OpenAI's ChatGPT, turning normal photos into cozy animated stills that look like they were pulled out of Hayao Miyazaki's masterpieces. In fact, so many of these images were made that many think they infected future AI images with their warm tones."
"But these images don't exist in a vacuum: in order to copy an anime style, these programs had to be trained on existing images - images that, intellectually, belong to others. Now, the Content Overseas Distribution Association, the industry body that represents several anime studios, including Toei and Square Enix, has sent a written request to OpenAI requesting that its members' products not be used to train its Sora program without permission."
Generative AI images developed a recognizable "AI look" characterized by flat lighting, nonsense text, and a yellowish tint. A large 2025 trend to convert photos into Studio Ghibli–style images using OpenAI tools produced many warm-toned outputs that likely influenced subsequent model outputs. Those Ghibli-style edits were created by training models on existing anime imagery, much of which belongs to studios. The Content Overseas Distribution Association representing studios such as Toei and Square Enix has formally asked OpenAI not to use members' works to train the Sora model. Japanese authorities and additional companies have requested opt-outs to protect copyrighted anime.
Read at Inverse
Unable to calculate read time
Collection
[
|
...
]