India’s public sector broadcaster Doordarshan recently introduced two AI anchors – Krish and Bhoomi – who deliver weather forecasts, commodity prices, farming trends, updates on agricultural research, and information on state welfare programmes to millions of farmers.
Enabling this is a Delhi-based startup called Personate AI. Incorporated in 2021, the startup is helping broadcasters, media houses and even content creators develop virtual AI agents.
Last year, Aaj Tak became the first broadcaster in India to host an AI anchor named Sana. The virtual anchor is multilingual and provides new updates multiple times throughout the day.
“We introduced India’s first AI anchor, Sana, at the India Today Conclave in the presence of Prime Minister Narendra Modi. Following that, we launched several anchors for various brands, including Vendhar TV in South India and ongoing campaigns for Zee.
“Since then, AI anchors have been developed for multiple news channels, including Russian media,” Rishab Sharma, chief technology officer & co-founder at Personate.ai, told AIM.
The startup has developed eight different AI anchors for Aaj Tak and successfully launched them across its regional channels. Last month, Modi’s interview with the channel was also translated and broadcasted in seven hyperlocal languages using the startup’s technology.
Building on this success, the startup approached Prasar Bharati, which was impressed with its vision and chose to integrate AI anchors for DD Kisan. Sharma also revealed that AI anchors will soon be introduced for other Doordarshan channels, including DD Sports and DD News.
Personate.ai is also co-founded by Akshay Sharma, who serves as the CEO. Completely bootstrapped and profitable, the company has over 25 enterprise customers.
Personate AI Studio
The startup has created an AI studio that allows users to produce a clone or a digital avatar of themselves. To date, the startup has collaborated with five media houses, achieving an average return on investment (ROI) of approximately 160%, according to Sharma.
“It’s adding more viewers per minute. Indian viewers have become accustomed to viewing AI content, be it a reel or a video shot on social media,” he said.
Personnate’s AI studio replaces the human component with a synthetic one. This synthetic element can be an actual human, or content creator or a synthetic personality.
For Aaj Tak, the startup also created a clone for their managing editor Anjana Om Kashyap. This was done by taking a short video clip, typically five minutes of the individual’s data. With AI, the clone was ready within minutes to read anything on screen with text-to-speech technology.
In contrast, creating a synthetic anchor involves designing from scratch. “This process includes pixel manipulation, designing the body, overlaying textures and clothing, and stitching the face. For a synthetic person, we spend about a week crafting the design,” Sharma revealed.
The personality is created with a 3D model by adding textures and shapes to the body. Once the model is ready, AI steps in to control its movements.
Explaining in the context of video games, Sharma said that traditionally animators decided how the character moved. “Here, the role of the animator shifts to generative AI, which now directs how the character behaves and interacts,” Sharma said.
Large Vision Models
Personate has developed a large vision model (LVM), which is a generative AI model similar to LLMs, but generates pixels instead of text. Examples of popular LVM include OpenAI’s Sora and Google’s Imagen.
The AI model translates a 3D anchor to a 2D screen, ensuring that the output is ultra-realistic. On a 2D screen, the challenge is how to rotate the model and adjust its positioning relative to camera angles, even though there is no actual camera.
“The goal is for the model to behave as if it’s responding to the camera, creating a convincing and lifelike appearance,” Sharma said.
One of the biggest challenges in training an LVM is data. Moreover, models like OpenAI’s Sora are trained on trillions of data points. According to Sharma, Personate’s AI model too is trained with multi-trillion data points.
The startup tapped into the past experiences of the founders to collect data to train the model. Sharma revealed he began his journey with the Indian Space Research Organisation (ISRO) and later worked with Reliance.
Currently, Personate.ai is the only Indian company with such capabilities. Beyond India, Synthesia, a startup based in London, offers similar solutions.
Synthesia’s platform enables users to create videos using pre-generated AI avatars or by generating digital representations of themselves, which they call artificial reality identities. The startup is backed by NVIDIA and is being leveraged by the United Nations.
