Cheaper, faster and culturally aware, Avataar’s video AI is designed for India’s scale

AI model production in India has been slow compared to the US, Europe and China. Only a few startups release models, and most of them are large language models or audio models. To encourage further development, the government launched the India AI Mission, a roughly $1.2 billion initiative that gives selected startups — among other things — access to GPU-supported computing in exchange for publicly launching their models. One of the 12 startups selected for the program, Avataar AI, has launched a new video model called Varya that is designed to understand local context – such as identifying different festivals, food, and clothing.

The Peak XV-backed startup, which focuses on creating e-commerce video tools, didn’t build Varya from scratch. It started with Wan 2.2, a publicly available video generation model released by Alibaba, and used a technique called distillation — essentially compressing the model’s capabilities into a leaner, faster version that was optimized for Avataar’s specific use cases. The result is a model that plays in four steps instead of 50 for the Wan 2.2, producing video 10 times faster at a fraction of the cost.

To put this in concrete terms: Using an NVIDIA H200 GPU, the Varya can create a 5-second clip at 720p in 45 seconds, compared to 1230 seconds for the Wan 2.2.

Perhaps the most striking aspect of the Varya is its price. The company plans to charge INR 0.48 ($0.005) per second for video on its hosted service — much cheaper than models like Veo, Kling, Luma and Runway, which typically charge $0.10 or more per second. That’s almost a 20x price difference.

“India is a video-first market. We see this across every major consumer internet product in India: video is trumping text. Current AI video models are too expensive for population-wide use in India. If video AI is going to reach students, educators, SMEs, creatives, enterprises and public services, costs will have to come down dramatically. Cost is the biggest lever for AI adoption in India,” Rajan Anandan, Managing Director, Peak XV, told TechCrunch.

Image and video generation models often miss cultural nuances and produce stereotypical or generic output — a problem TechCrunch has previously reported on. Avataar AI says it used curated data to train Varya to recognize cultural nuances including food, clothing, architecture and festivals.

Varya will be released as an open-weight model on the Indian AI Kosh portal — the Indian government’s central repository for publicly available AI models and datasets — along with its training data, meaning developers can self-host it or modify it according to their own needs. Avataar also plans to make the model available to its enterprise customers and says it is open to partnerships with video tools including Higgsfield and Adobe Firefly. Anyone can try it now on their website using text prompts or reference images.

The launch of Varia reflects a fundamental trade-off in India’s AI ambitions. Industry experts have observed that India can make its mark in the AI ​​space by creating applications and a strong developer ecosystem rather than competing on basic models. There is a reason for this pragmatism: model development has been slower in India than in global competitors due to lack of computing and limited availability of high-quality data.

India’s AI mission is also part of a broader government drive to bridge this gap. Last year, it selected 12 startups – including Avataar AI – to develop AI models and power them with cost-effective computing. Earlier this year, IT Minister Ashwini Vaishnaw said India aims to attract $200 billion in investments in artificial intelligence by 2028 and more than double its graphics processing capacity within six months.