In New Delhi, Avataar.ai unveiled Varya, a distilled video-generation model tailored for India and developed with support from the IndiaAI Mission to scale practical AI tools for the mass market.

The government-backed Varya aims to be faster and cheaper than leading Chinese models, including Alibaba’s Wan, with cost-conscious pricing and accessible performance.

Varya’s training emphasizes Indian cultural accuracy, focusing on clothing, architecture, festivals, and culinary traditions to reflect regional diversity.

The rollout prioritizes practical, scalable video AI solutions for students, teachers, SMEs, creators, and public services, rather than only core model development.

India’s launch reflects a pragmatic AI strategy: emphasize practical applications and a strong developer ecosystem over core-model competition due to limited compute and data quality.

Varya will be accessible via text prompts or reference images, with potential integrations into broader tools like Adobe Firefly, signaling an ecosystem approach.

Varya is distilled from a larger Wan 2.2 teacher model, cutting generation steps from 50 to four and delivering roughly tenfold speed gains over its predecessor.

Affordability is positioned as essential for widespread AI-driven video adoption in India, not merely a feature.

The India AI Mission, a government initiative around $1.2 billion, subsidizes GPU compute for select startups in exchange for public release of their models.

Varya’s development and access to low-cost GPU compute were supported by the IndiaAI Mission, signaling strategic alignment with national AI goals.

The broader program seeks to accelerate AI adoption by building a robust developer ecosystem and practical applications, aligning with goals to boost AI investment and expand GPU capacity by 2028 and beyond.

India intends to build indigenous AI capabilities beyond large models, focusing on affordable, scalable tools for population-wide deployment across education, commerce, and public services.