Amid government support to boost AI adoption, Avataar AI has introduced Varya, a distilled video generation model tailored to India's unique cultural landscape. The model promises faster output and dramatically lower costs compared to existing global alternatives.
- Varya generates video 10x faster and at 1/20th the cost of comparable models
- Model trained to identify cultural nuances like festivals and clothing specific to India
- Available as an open-weight model on India’s AI Kosh portal for public use
What happened
Avataar AI, supported by the India AI Mission and backed by Peak XV, has launched Varya, a video AI model that produces content rapidly at a fraction of the usual cost. By leveraging distillation techniques on Alibaba’s public Wan 2.2 video generation model, Avataar cut processing steps from 50 to 4, enabling the production of a 5-second 720p video clip in 45 seconds using NVIDIA H200 GPUs.
The startup designed Varya specifically for the Indian market, incorporating training data that focuses on cultural elements like festivals, food, and regional clothing. This approach addresses common challenges where AI models tend to overlook or generalize culturally specific details. Varya is accessible via Avataar’s website and will be available as an open-weight model on India’s AI Kosh platform, encouraging community-driven development and customization.
Why it matters
India is a predominantly video-first market, with video content favored over text in many digital interactions. However, high costs and slow video AI processing have hindered broad AI adoption. Varya’s low pricing at approximately $0.005 per generated second—around 20 times cheaper than competitors—could significantly lower the barrier for widespread video AI use among enterprises, content creators, educators, and MSMEs across India.
Moreover, the model’s cultural awareness represents a crucial advance for AI in India, helping to diversify application outputs and reduce stereotypical or generic content. This emphasis resonates with ongoing national efforts under the India AI Mission to expand India’s AI capabilities by providing startups access to subsidized compute and encouraging the release of publicly accessible AI models tailored to local needs.
What to watch next
Going forward, the success of Varya will depend on its adoption by Indian developers and enterprises, as well as its integration with major video editing and content platforms like Adobe Firefly. Collaborative efforts with partners such as Higgsfield could expand its reach and functionality beyond initial use cases in e-commerce video production.
On a broader scale, India’s government initiatives aiming for a $200 billion AI investment influx by 2028 and significant increases in GPU capacity will be critical. These moves are expected to accelerate the development of more specialized models like Varya and stimulate a vibrant local AI ecosystem focused on practical applications rather than foundational model competitions.