S Krishnan, Secretary, Ministry of Electronics and Information Technology |Photo Credit: SUSHIL KUMAR VERMA
Backed by IndiaAI Mission, artificial intelligence (AI) native transformation company Avataar.ai on Friday launched Varya, a distilled video model, which it said will make frontier video AI affordable, accessible and relevant for India’s next generation of users.
Speaking at the launch, S. Krishnan, Secretary, Ministry of Electronics and Information Technology, said that India is not lagging behind in efforts to develop basic AI models and the models developed under the India AI Mission are an answer to those who are doubting the country’s potential in developing advanced AI platforms.
“For all the people who say that model building and this type of initiative won’t work in India, this (Varya) is the answer to show that it is actually possible and the answer to show that India is second to none in this type of initiative,” he said.
Krishnan added that the government supports not just language models, but a wide range of underlying models across a variety of applications.
Distilled video generation in machine learning is a model compression technique where a compact “student” model replicates the output of a larger, slower “teacher” model, thereby transferring functionality while eliminating redundant computations.
While a standard video model must iterate over 50 noisy steps before producing a clean output, distillation teaches the Student model to skip most of these steps and provides the same quality in just a few iterations.
Avatar said Varya will be the first in India to apply this breakthrough to video, making frontier quality video generation affordable, fast and at population scale.
According to Sravanth Arul, chief executive officer and co-founder of Avatar, Varya is not a generalized concept of India but is designed to suit different Indian contexts. The model is built to understand and generate culturally rich visual output across India’s regions, festivals, communities, food, clothing, public spaces and daily life, it said, adding that users can create 211 seconds of video for every ₹100 spent on the platform.
He claimed that Varya reduces video generation time from 50 steps to 4 steps, making it 10 times more efficient than leading models. He added that the company has developed a video AI platform with 14 billion parameters.
“From teachers creating visual lessons in village classrooms, to micro, small and medium enterprises (MSMEs) creating product advertisements, to citizens accessing public information through video, Varya aims to transform simple ideas into inspiring stories,” he said.
Avataar also plans to publish a technical report outlining Varya’s model architecture, distillation techniques, and benchmarks.
Published June 12, 2026
