A research team from Nanjing University's MCG-NJU and Tencent PCG has released SteadyDancer, a framework that generates human animation video from a single reference image and a pose-driving video. It is designed to suppress identity drift, where a person's appearance degrades during generation, and spatio-temporal misalignment caused by discrepancies between the reference image and the driving video. The paper (arXiv:2511.19320), code, model weights, and an evaluation benchmark were released on November 24, 2025.
Continue reading
The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in✓ Signed in — this article isn’t included in your current plan.