Hugging Face hosts models that easily create deepfake nudes, study finds

Popular image editing models on Hugging Face can be easily manipulated to create explicit deepfake nudes, a study released on 28 July 2026 found. Researchers also uncovered more than 1,000 image editing prompts that show how people misuse the software.

The tested models rank among the platform’s most popular, and the prompts were collected from public sources, indicating a systemic problem. No specific models or users were named.

Hugging Face has previously introduced safety features, but the report suggests they are insufficient to prevent non-consensual intimate imagery. Without intervention, proliferation of such content could cause severe psychological harm, reputational damage and an erosion of trust in digital media.

To curb misuse, the researchers recommend immediate measures: mandatory watermarking, stricter access controls and enhanced detection algorithms. They also urge collaboration among platforms, policymakers and AI developers to establish enforceable standards.

The study’s release is expected to prompt a review of Hugging Face’s content policies and could lead to the removal or modification of vulnerable models. The findings are likely to reignite debate over the responsibility of AI repositories in preventing harm.