Hollie Mengert art used to train Illustration Diffusion

Occurred: November 2022

Can you improve this page?
Share your insights with us

Illustrator Hollie Mengert discovered that her online art portfolio was used to train Illustration Diffusion, a text-to-image model created by Canada-based Nigerian engineering student Ogbogu Kalu, without her permission.

Per Andy Biao at Waxy, Kalu used 32 of her illustrations to fine-tune Stable Diffusion to recreate Hollie Mengert's style using Google's DreamBooth, a technique for introducing new subjects to a pretrained text-to-image diffusion model. Kalu then released the model on Hugging Face under an open license for anyone to use.

The act triggered a heated debate about the ethics and legality of using artwork developed and owned by other people or organisations without their consent. Dreambooth was also criticised for the ease with which it can be used to generate offensive or malicious images, and that it can be re-purposed given its open source nature.

Mengert pointed out to Andy Baio that she was in no position to grant Kalu permission to train his model on her work even if she wanted to as her work involves characters owned by corporations like Disney or Penguin Random House.

On the other hand, Kalu said he thinks his act was legal and 'likely to be determined fair use in court'. He reckoned it is also inevitable. 'The technology is here, like we've seen countless times throughout history,' he argued. According to Kalu, 'there is no argument based on morality. That's just an arbitrary line drawn on the sand. I don't really care if you think this is right or wrong.'

Operator: Ogbogu Kalu
Developer: Ogbogu Kalu; Alphabet/Google

Country: USA

Sector: Media/entertainment/sports/arts 

Purpose: Fine-tune text-to-image models

Technology: Text-to-image; Machine learning
Issue: Copyright; Ethics; Mis/disinformation; Safety

Transparency: Governance

Page info
Type: Incident
Published: February 2023