Aespa Karina deepfake (Jun 2025)

In June 2025, Waterbomb Seoul, one of South Korea's largest summer music festivals, drew intense criticism after AI-generated imagery of aespa's Karina appeared in its official promotional material without her consent. Unlike most deepfake incidents involving K-pop idols, this one came not from anonymous abusers but from a commercial event that had booked the artist to perform.

What happened

On June 10, 2025, Waterbomb announced the lineup for its 2025 Seoul festival, with Karina among the performers alongside acts such as ITZY and Jeon Somi. In the accompanying teaser video, a still photograph of Karina had been animated with AI so that she appears to blink, look around, and move — footage she never filmed — as reported by Koreaboo and the Times of India. Fans identified the synthetic footage almost immediately and objected, arguing that normalizing AI recreations of an idol's likeness opens the door to more harmful misuse. Criticism landed both on the festival's organizers for producing the material and on SM Entertainment for failing to prevent an unauthorized AI depiction of its artist.

The backlash drew force from context. aespa's members have been repeated targets of non-consensual deepfake material, and SM Entertainment has publicly committed to responding firmly to synthetic-media abuse of its artists. In 2026, a Seoul court sentenced a producer and seller of deepfake content targeting aespa to two years and six months in prison, according to The Kyunghyang Shinmun.

Why this incident matters

Most deepfake coverage focuses on fraud or explicit abuse, but this incident sits in a different category: commercial use of a real person's AI-generated likeness without consent. Even a seemingly harmless animation — a blink, a turned head — puts an artist's face into motion she never performed, and doing so in official marketing normalizes the idea that a photograph is raw material for synthesis. For an industry whose performers are disproportionately targeted by malicious deepfakes, that normalization carries real cost, and it exposes a gap in likeness-rights enforcement that persists even for artists backed by major agencies.

Where detection fits

AI-animated stills are difficult to spot because most frames stay close to an authentic source photograph. An image detector can flag generated frames, since content produced by a diffusion model carries statistical fingerprints that survive even subtle animation. For agencies and rights holders, automated screening of promotional material offers a practical checkpoint before an unauthorized likeness reaches the public.