India Launches Varya as First Indigenous Video AI Model

Electronics and IT Secretary S Krishnan launched Varya, India's first indigenous video foundational AI model, developed by the startup Avataar.ai. Supported by the IndiaAI Mission, Varya utilizes 14 billion parameters and allows users to generate edited videos from photos and raw footage via text prompts. The model employs distilled video generation to reduce computation steps from 50 to four, resulting in a generation speed and cost reduction 27 times greater than leading competitors, with costs reaching Rs 0.48 per second.

Varya is one of five models already released under a broader government initiative to create 20 foundational AI models. Other recent releases include a 105 billion parameter model from the startup Sarvam and a multimodal large language model from the BharatGen academic consortium that supports 22 Indian languages. To sustain this sovereign AI development, the Ministry of Electronics and Information Technology provides funding for research consortia and subsidizes access to over 38,000 graphics processing units.

Avataar.ai CEO Sravanth Aluru stated that the model aims to unlock productivity in India through affordable AI and will open markets that were previously underserved. While the tool automates repetitive manual work, Aluru emphasized that human ingenuity remains essential, highlighting a need for workers in the video development segment to upskill.