IndiaAI Mission backs 'Varya', an indigenous, India-context video-generation AI model

Built by Avataar.ai using subsidised national AI compute, Varya is a distilled text-to-video model designed for India's languages and contexts, claiming far greater efficiency than leading models.

What happened

• Under the IndiaAI Mission, the AI-native company Avataar launched 'Varya', a distilled video-generation AI model built to make frontier video AI affordable, accessible and relevant for India's next generation of users.
• It was unveiled in New Delhi in the presence of Shri S. Krishnan, Secretary, MeitY, alongside Avataar's leadership.
• Varya is designed for India's many contexts — generating culturally rich visual outputs across India's regions, festivals, communities, food, clothing, public spaces and everyday life.
• Avataar was among the companies selected by the IndiaAI Mission to build indigenous foundation AI capabilities; access to subsidised national AI compute infrastructure enabled the research that led to Varya — showing how public AI infrastructure can accelerate homegrown innovation.
• Varya claims to cut video-generation time from 50 steps to 4 (about ten times more efficient) than leading models, with applications from a teacher's village classroom lesson to an MSME's product ad to citizen access to public information through video.

For Prelims

• IndiaAI Mission: Approved in March 2024 with an outlay of ~₹10,371 crore (over five years), implemented by MeitY (via the IndiaAI division of Digital India Corporation). It has seven pillars — including IndiaAI Compute, Innovation Centre, Datasets, FutureSkills, Startup Financing and Safe & Trusted AI.
• IndiaAI Compute: The pillar building shared AI computing infrastructure (deploying 10,000+ GPUs via public-private partnership) and offering subsidised compute to startups/researchers — the enabler behind Varya.
• Foundation / generative models: Large AI models trained on vast data that can generate text, images or video. Varya is a text-to-video generative model; 'distilled' means a smaller, efficient model derived from a larger one.
• Why 'India-context': Most frontier models are trained on Western data; India-context models aim to represent India's languages, festivals and visual culture — complementing multilingual efforts like Bhashini.
• Indigenous foundation models: The IndiaAI Innovation Centre's Call for Proposals (from 2025) backs Indian firms (e.g. Sarvam, others) to build sovereign foundation models — Varya is part of this push for AI self-reliance.
• Safe & Trusted AI: An IndiaAI pillar addressing bias, deepfakes and safety — relevant given video-generation models' deepfake risks.
• Don't confuse: Varya is an industry-built model supported by the IndiaAI Mission (public compute + selection) — not a government-built product; and a 'distilled' model is an efficiency-optimised derivative, not a brand-new architecture from scratch.

For Mains

Syllabus: GS3.13 · GS3.11 · Linkage L2