A team from Alibaba has launched a new video generation tool called Tora020 Archives emphasizing its “excellence” in achieving high-motion fidelity, as buzz around OpenAI’s text-to-video generator Sora drives Chinese companies to develop similar AI models. In dozens of video clips posted by the team and produced via Tora, the model’s trajectory-oriented diffusion transformer for video generation ensures that the generated motion follows the specified trajectory drawn in advance. According to the Tora introduction paper, the team said Sora-like models are limited in “effectively generating videos with controllable motion,” while claiming that Tora is the first trajectory-oriented diffusion transformer framework. The work has not yet been finalized, said the paper, noting that Tora would be added to the official Alibaba repository once it has received company approval. [Tora]
