Video developers can combine visual references and a voice reference while evaluating latency across different hardware recipes. Core capabilities include Prompt-conditioned video and audio, Image-reference input, Video-reference input, and Voice-reference conditioning.
Public code or model weights are linked from the project, with setup instructions for running or adapting the system. Additional characteristics include Eight transformer evaluations, UniServe deployment, NVFP4 QAD weights, and ComfyUI support.
