Step-Video-T2V: Text-to-Video Generation Model by StepFun-AI

The provided texts focus on Step-Video-T2V, a 30-billion-parameter text-to-video model.The model uses a deep compression VAE to enhance efficiency and DPO to improve video quality, achieving state-of-the-art results. The texts mention the release of code, model weights, a technical report, and a new benchmark for video generation quality, Step-Video-T2V-Eval.This includes instructions for model download and usage, covering hardware and software requirements and optimal inference settings. Furthermore, the online version and citation information were released.Sources:​Stepping-ai : GitHub ​Step video t2v : Huggingface