Early Failure Detection and Intervention in Video Diffusion Models

Kwon Byung-Ki, Sohwi Lim, Nam Hyeon-Woo +2

Text-to-video (T2V) diffusion models have rapidly advanced, yet generations still occasionally fail in practice, such as low text-video alignment or low perceptual quality. Since d…