Detect Deepfakesby Resemble AI
Deepfake case study · Video

Game of Thrones deepfake (Feb 2026)

An AI-generated alternate Game of Thrones ending made with ByteDance's Seedance 2.0 went viral, drawing objections from Warner Bros. and SAG-AFTRA.

Incident date
Feb 2026
Target
Game of Thrones
Updated May 6, 2026 · 2 min read

In mid-February 2026, within days of ByteDance releasing its Seedance 2.0 video-generation model, AI-made clips rewriting the ending of Game of Thrones began spreading across social platforms. In the fan-made "fix," Daenerys Targaryen never descends into madness — she survives the finale and rules alongside Jon Snow. The clips used the likenesses of the show's real actors and HBO's copyrighted characters without permission, and they were convincing enough that some viewers judged the effects favorably against the original series.

What happened

Seedance 2.0 launched in February 2026 and users immediately began generating unauthorized videos built on famous film and television properties. Alongside the Game of Thrones alternate ending, viral examples included a staged fight between Tom Cruise and Brad Pitt and a clip of Rocky Balboa meeting Optimus Prime, according to The Hollywood Reporter. Forbes reported that fan-made "better endings" for Game of Thrones and Stranger Things went viral and described the renders as the most convincing to date.

The backlash was swift. Warner Bros. Discovery, whose HBO unit owns Game of Thrones, publicly objected to AI videos of its characters, as reported by Variety. Motion Picture Association CEO Charles Rivkin said the tool enabled "unauthorized use of U.S. copyrighted works on a massive scale" and called on ByteDance to stop. SAG-AFTRA, the union representing screen actors, condemned what it called "blatant infringement" involving the unauthorized use of members' voices and likenesses. Forbes reported that Disney also demanded removal of infringing videos generated with the tool.

Why this incident matters

The Seedance episode crystallized a growing conflict between fan-made synthetic media and intellectual property rights. Even a "harmless" alternate ending depends on replicating the faces and voices of real performers who never consented, and on characters a studio spent hundreds of millions of dollars creating. As video models make studio-grade fan fiction trivial to produce, rights holders, unions, and platforms are being forced to decide where homage ends and infringement begins — and actors face a future in which their screen likeness can be recast in scenes they never performed.

Where detection fits

As AI-generated scenes of real actors become harder to distinguish from broadcast footage, verification matters for platforms, rights holders, and viewers alike. A video deepfake detector can flag synthetic footage of known performers before it spreads as authentic, and understanding techniques like face swap helps explain how a familiar actor's likeness ends up in scenes that were never filmed.

Sources