Japanese developers are taking deepfake to the next level through an AI system that can create hyperrealistic full-body images of people who do not exist in this world. Data Grid, the Japanese artificial intelligence development firm behind the new technology reportedly used massive amounts of data to create hyperrealistic images.
To date, the images generated by the Japanese AI are the most realistic experts have encountered. The AI technology is reportedly developed for clothing companies and advertising agencies who want to employ photogenic models for their brands without the extra cost.
In the past, Data Grid was able to develop an AI that also auto-generates images. However, the system’s capability was only limited to producing face images, therefore lacking expressive power.
Fake Full-Body Images of People
In a press release posted on Data Grid’s website, the company said:
“Datagrid developed idle automatic generation AI in June 2018. However, idle auto-generation AI generates an image only for the face area, so it did not have enough expressive power.
Therefore, in order to enhance the expressive power of the generated person, we have been working on two research and development of “whole body generation” and “motion generation”.
The high-precision whole-body generation model has no precedent and was a challenging research and development. We have succeeded in stably generating high-resolution (1024 × 1024) whole-body model images.”
Data Grid’s AI system uses a Generative Adversarial Network algorithm. It is the kind of artificial intelligence software that creates imitations by utilizing the data it gathered of objects that exist in the real world.
Previous GAN efforts were able to generate images of people but with noticeable flaws. For instance, some faces created by AI GAN tend to be asymmetrical or have blotches bleeding out from the background.
Data Grid was able to resolve this problem by making the AI build its hyperrealistic models in front of a nondescript white background, with equally realistic light shining down on the models. According to the company, they are still working on perfecting the AI’s motion generation capability and will soon demonstrate their new technology to potential stakeholders.
Comments (0)
Most Recent