Microsoft tweaks AI image generator over Disney dogs poster trend

Stay informed with free updates

Simply sign up to the Artificial intelligence myFT Digest — delivered directly to your inbox.

Microsoft has tweaked its artificial intelligence image generator tool after concerns were raised over a viral social media trend where users created realistic Disney film posters of their dogs, highlighting broader copyright issues in the industry.

Disney’s logo was visible in illustrations made by Microsoft’s Bing Image Creator, which were then posted to TikTok and Instagram. The pictures featured people’s pets in the style of posters made by Disney’s Pixar studio, leading influencers to encourage others to prompt the artificial intelligence tool to generate their own versions.

After concerns were raised, the term “Disney” was blocked from being entered into the image generator, and users were shown a screen which said the prompt — the search terms used to guide the AI — was against its policies. One person familiar with Disney’s approach suggested the media giant had reported its concerns about copyright or intellectual property infringement.

The tool has since been adjusted to allow the term Disney, but the text and logo in the AI-generated pictures still resemble those in original images, as illustrated in the poster, above, for the film Brave. They feature a jumbled but still recognisably similar version of the genuine logo but, for example, with misspellings or a blurred font, such as the picture above showing a dog.

“Reproduction of the Disney logo is clear trademark infringement. I would imagine that is why the AI might be jumbling the logo,” said Andrew White, partner at IP law firm Mathys & Squire.

He added there was an “unresolved issue” over whether the AI models had been trained on Disney’s content and also if it was reproducing copyrighted material.

“In this instance, it is likely more for fun than trying to pass the images off for their own film or animation, but that would be a huge issue. From a PR perspective, it is a fine line for Disney seeming to come down on something for fun that is ultimately raising its profile versus something that is actually anti-competitive,” White added.

In September, Microsoft pledged to assume legal responsibility for any commercial customers using its software in Word, PowerPoint and coding tools who face complaints of breach of copyright due to material generated by its AI products. Other big tech firms using generative AI, including Adobe and OpenAI, have offered similar protections.

Influencer accounts on social media were instructing followers to input terms such as “a Disney Pixar-inspired movie poster”, which would create images that used the media giant’s logo, among other design characteristics.

Nicola Bennett, who runs an Instagram account for her Italian Greyhound under the handle Pandora the Iggy, said she made images after seeing others do it on the social media platform.

“I wrote in Disney-style cute poster — the images are so cute and similar, I can’t believe it makes them so fast,” she said.

Bennett said she thought it was “a positive thing for the Disney brand, but maybe they should have their own generator”. She also recognised that “now you can do it free . . . that can take away from people who would make it by hand”.

Microsoft would not comment on commercial discussions but said: “There is a current level of variability that may return different results from time to time as we continue to refine our safety systems . . . Additionally, artists, celebrities, and organisations can request to limit the creation of images associated with their names and brands.”

Disney declined to comment.

The incident comes as artists, singers, media companies and publishers have claimed that copyrighted materials have been used to develop AI products without their consent or payment.

Several lawsuits are under way amid questions over what data the models underlying the technology have been trained on. Stock image provider Getty Images is suing Stability AI, alleging the company used its images to train its model.

Additional reporting by Christopher Grimes in Los Angeles