Researchers from Peking University have Future Sex (2018)formed a small team that is aiming to “reproduce” OpenAI’s text-to-video generator Sora, according to its project page on GitHub, after the original model took GenAI to the next level following its roll out in February. Collaborating with Shenzhen-based AI firm RabbitPre, China’s top academic institution is seeking resources from the open-source community as they are “unable to conduct full training” due to a lack of video data. However, the team has developed a three-part framework – including a Video VQ-VAE, a Denoising Diffusion Transformer, and a Condition Encoder – and presented its initial progress on GitHub, with four reconstructed video demos offering different resolutions and aspect ratios, ranging from three to 24 seconds in length. [GitHub]
