Скачать или смотреть EPFL Deepfake Booth AI For Good 2025

EPFL’s Deepfake Booth at the AI for Good Summit 2025 in Geneva invites visitors to witness how generative adversarial networks can fabricate persuasive audiovisual fabrications from a single selfie. After a quick portrait capture, attendees step into a darkened cabin where a one-minute primer outlines social and political risks associated with synthetic media. The installation’s goal is literacy rather than spectacle, underscoring that trust must be earned, not assumed. https://www.epfl.ch

Inside the booth a lightweight face-swap pipeline blends an encoder–decoder front end with a diffusion-based refinement stage, rendering photorealistic talking-head or composited stills in under sixty seconds flat. The GPU-accelerated stack exploits identity embeddings, landmark alignment and real-time style transfer so that users instantly see themselves delivering fabricated statements or even “standing” in a prison mug-shot. Because the template clips and backdrops are parameterizable, the platform doubles as a pedagogical sandbox for exploring face-reenactment, latent-space interpolation and adversarial robustness testing.

EPFL researchers leverage the same venue to preview countermeasures. The Image & Visual Representation Laboratory (IVRL) recently published an “Efficient Temporally-Aware DeepFake Detection using H.264 Motion Vectors” framework that flags frame-to-frame geometry inconsistencies without expensive optical-flow decoders, enabling low-cost forensic scans of streaming video. Parallel projects probe codec-level watermarking, multimodal provenance metadata and frequency-domain artefact localisation, illustrating how detection techniques must evolve alongside generative models.

That arms race also surfaced in an ITU report released during the summit, which called for global standards so platforms can authenticate multimedia before distribution, warning that manipulated clips already threaten elections and financial security. Against that backdrop EPFL hinted at a forthcoming Swiss LLM co-developed with ETH Zürich, slated for public release in late 2025, and urged visitors to triangulate sources, check hashes and question sensational footage to safeguard an information ecosystem fit for the future.

https://actu.epfl.ch/news/low-cost-de...
https://www.reuters.com/business/un-r...

My full AI For Good 2025 video coverage is here:    • AI For Good 2025 video coverage  

This video was filmed using the DJI Pocket 3 ($669 at https://amzn.to/4aMpKIC using the dual wireless DJI Mic 2 microphones with the DJI lapel microphone https://amzn.to/3XIj3l8 ), watch all my DJI Pocket 3 videos here    • Filmed with DJI Pocket 3  

Join https://www.youtube.com/charbax/join for Early Access to my videos and to support my work, or you can Click the "Super Thanks" button below the video to send a highlighted comment under the video! Brands I film are welcome to support my work in this way 😁

Check out my video with Daylight Computer about their revolutionary Sunlight Readable Transflective LCD Display for Healthy Learning:    • Daylight Computer CEO interview: Healthy C...