Visible artifacts are an unreliable cue for detecting generated videos, because newer generators learn to remove them. We instead detect a trace left by the generation pipeline itself. Most video generators decode latents with a 3D causal VAE that groups consecutive frames in time, so the fidelity o
VAE Rhythm: Temporal Decoding Traces for AI-Generated Video Detection
A new detection method called VAE Rhythm identifies AI-generated video by analyzing temporal decoding traces left by the 3D causal VAE that most video generators use to group consecutive frames, rather than relying on visible artifacts that newer generators learn to remove. The approach targets the fidelity patterns the generation pipeline itself imprints across frames, according to the research. The work matters because artifact-based detectors degrade as generators improve.
Build generated-video detection on the decoder's temporal grouping traces instead of visible artifacts, since newer generators erase artifact cues.
more in playbook: Evals and testing for LLM appsRun your AI side-project on zahid.host
EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.