Avataar.ai launches indigenous video model Varya to drive cost-efficient AI accessibility across India
New Delhi, June 12
Avataar.ai launched Varya on Friday, an indigenous, distilled video model designed to make frontier video artificial intelligence affordable and culturally relevant for users across India. The model, unveiled at a press event in New Delhi, aims to lower execution costs while accurately representing the country's regional and visual diversity.
According to an Avataar press release, the company developed the model to understand and generate visual outputs spanning India's distinct regions, festivals, communities, and public spaces. The initiative specifically targets applications in education, commerce, governance, citizen services, and digital storytelling for micro, small, and medium enterprises (MSMEs).
Avataar was selected by the IndiaAI Mission to build indigenous foundation AI capabilities. The research utilized subsidized national AI compute infrastructure to accelerate the development of the model.
Speaking at the launch, S. Krishnan, Secretary, Ministry of Electronics and Information Technology (MeitY), Government of India, said, "The launch of one of the foundational models supported under the IndiaAI Mission marks a significant milestone in India's AI journey. Varya represents the kind of research-led capability building that we seek to enable."
The technical framework of Varya relies on a distillation technique that reduces video generation from 50 steps to four steps. The company's internal inference-cost benchmarks indicate that the model "generates video at Rs 0.48 per second, which makes it up to 10 times more cost-efficient than several leading global video models."
"It is a proud moment that reflects our commitment to building indigenous AI capabilities and fostering a vibrant deep-tech ecosystem," Krishnan added. "Through strategic support for foundational models, we are enabling innovation at scale and creating the building blocks for the next generation of AI solutions. We look forward to further strengthening India's AI capabilities and advancing frugal innovation that is both world class and accessible at population scale."
Sravanth Aluru, CEO and Co-Founder of Avataar, said, "India's AI opportunity will not be defined only by the largest models. It will also be defined by the most efficient models. Varya demonstrates that frontier-quality video AI can be made dramatically more efficient and accessible. For a country of 1.4 billion people, affordability is not a feature, it is a prerequisite. We believe the next billion stories, lessons, advertisements, services and experiences will be created through AI, and those capabilities must be available to everyone, not just a few."
The application functions on a sequence where users input text or upload an image to generate a video clip, with the ability to extend the sequence through additional clips. Avataar stated that it will publish a technical report detailing the model's architecture, distillation methodology, and benchmarks.
Aluru said, "Affordable AI is inclusive AI. What Avataar is building with Varya is important because it shifts the conversation from AI capability alone to AI accessibility. India has the talent, the market depth and the imagination to build AI products for population-scale use," Aluru added. "Varya is a strong example of how Indian companies can compete globally, not by simply building bigger models, but by building more efficient, context-aware models that expand access."
— ANI
Reader Comments
Impressive technical achievement with the 50-step to 4-step distillation. But I hope they release a proper technical paper with benchmark comparisons against global models like Sora or Runway. We need transparency to ensure the 'cultural relevance' isn't just a marketing buzzword. Still, great to see IndiaAI Mission funds being put to good use!
As someone working in AI research in Bangalore, this is genuinely exciting. The cost efficiency (10x cheaper than global models) could democratize video creation for education and government services. I'm curious how well it handles low-resource Indian languages though. The real test will be when a village school teacher in Bihar tries generating a science video in Maithili.
'Affordability is not a feature, it is a prerequisite.' - Aluru's statement hits the nail on the head. For too long, AI tools were priced for Silicon Valley, not for Saket or Koramangala. Varya could help MSMEs create product demos, local ads, even wedding invitations with AI. But I hope they keep it open-source or at least provide an API for Indian developers. Our startup ecosystem needs accessible building blocks!
Does anyone know if Varya can generate videos in regional scripts for text overlays? For governance applications like public service announcements in rural areas, that would be critical. Also, I'm a bit concerned about the compute infrastructure - using subsidized national AI compute is great, but we need to ensure smaller players can also access it without bureaucratic hurdles. Nonetheless, proud of this indigenous achievement! 🚀
We welcome thoughtful discussions from our readers. Please keep comments respectful and on-topic.