Synthesia has partnered with Shutterstock to train its AI models as the text-to-video AI platform looks to launch a range of AI-driven avatars later this year.
Through the partnership, the Synthesia research and development team will have access to the video and image provider’s content library through a research licence with the option to extend it for commercial use.
Synthesia will review video data and use relevant content such as cinematic video and footage of individuals engaged in professional tasks, to enhance its existing data sets.
The material will be used to train Synthesia’s EXPRESS-2 model and create the new range of AI avatars.
Synthesia said that building a large AI model involves a lot of research-focused experimentation before it can be put into production, with the company testing various approaches.
As part of this experimentation, the company has paid “thousands” of actors and studied their performances to try to increase the realism and expressiveness of AI generated avatars.
“The new EXPRESS-2 model will represent a leap forward in AI-driven video and audio synthesis,” said Alexandru Voica, head of corporate affairs and policy at Synthesia. “With the enriched training data from Shutterstock, we hope to accelerate our research and produce more capable models that deliver performances with nuanced expressions and gestures that closely mimic human behaviour.”
