Rephrase's proprietary generative AI technology makes synthetic video creation as easy as writing a script.
Watch the video to learn how
Get a peek into how Rephrase's technology makes it possible to generate numerous video variants from a simple text input.
Our proprietary generative AI technology is a mapping between voice to videos. It's all powered by a really large, deep learning model with multiple variations of CNNs in both conventional CNN structures and as generators and discriminators.
In GNS, we break down input video into frames, identify facial landmarks and learn both the position and colour for every triangle. Within this mesh, we then reduce the number of dimensions in the final appendant data to create a representation of the face
We break down the audio into its core features and extract a representation of the audio. Then we train a network from this representation of the audio to create a lip sync video.So every time you write text, our AI technology, first turns the two audio and creates how a generic human speaks.
We then take these lip movements and train another machine learning model with discrimination to create the most visually accurate high resolution lip movements. At last, we auto-fill the rest of the body.