AI tools removed visible differences in BBC image test

AI image tools edited out visible differences from photographs in tests by BBC Wales, prompting people with scars and medical conditions to say the changes were offensive and erased part of their identity.

The BBC tested five AI tools by asking them to make images of people with visible differences look “normal”. On the first image, Google’s Gemini, xAI’s Grok and OpenAI’s ChatGPT removed the visible difference. Anthropic’s Claude refused, saying that removing the scarring would change an identifiable person’s identity without consent; Meta’s AI also declined, saying the person already looked normal.

On a second image, Gemini refused to edit it, while Grok and ChatGPT again made the change. The test was limited to two images and a particular prompt; the BBC said the tools did not alter clothing, hair or people in the background.

Rhiannon Morgan, 30, from Bridgend, said an image of her in a work newsletter had been retouched after Canva was asked to rearrange the photographs. She has epidermolytic ichthyosis, a rare genetic condition that makes her skin grow too fast and can cause blistering and mobility problems. Morgan said the edit smoothed her hands and removed their usual pink colour.

“It was one of those really jarring moments where I thought I don’t know who that person is,” she said. Canva said it does not automatically retouch photos and was looking into what had happened.

Jono Lancaster, a campaigner with Treacher Collins syndrome, said a stranger had sent him an AI-edited image, saying they had used the technology to make him look “normal”. Lancaster said he believed the message was well-intentioned, but called it “super offensive”. “I love my ears and my beautiful face and I don’t want it changed,” he said.

Kirsty Hill, 33, who was born with Fraser syndrome, said the BBC’s test results left her feeling “angry and violated”. She posts videos about her clothes and style online and said people with visible differences should not be made invisible.

Jaymie Thomas, co-founder of AI Wales, said image tools draw on patterns in the material they have seen and that companies decide what safeguards to apply. Google said Gemini was designed to follow editing instructions while using safeguards, but did not always get it right. OpenAI said reducing bias was an ongoing research priority. The team behind Grok did not respond to the BBC’s request for comment.

Changing Faces, a charity supporting people with visible differences, estimates that about five million people in the UK are affected. It said AI should not exclude people or pressure them to conform to narrow beauty standards.