Avataar’s Varya: Cheaper, Faster Video AI for India's Market
India is taking steps to bridge the AI gap with Avataar’s new model, Varya. Find out how it aims to revolutionize video generation in a cost-effective way.
Admin User
⚔️ Specs Battle Simulator
Comparing rumored/leaked specs of Varya Video AI Model against the official Wan 2.2 Video Generation Model by Alibaba
| Specification | Wan 2.2 Video Generation Model by Alibaba (Official) | Varya Video AI Model (Rumored) | Difference |
|---|---|---|---|
| Speed of Video Generation (5-second, 720p clip) | 1,230 seconds with Wan 2.2 | 45 seconds with Varya | Upgrade 🔥 |
| Cost per Second of Video Generation | $0.10 or more for other models like Veo, Kling, Luma, and Runway | ₹0.48 ($0.005) for Varya | Upgrade 🔥 |
India has long been lagging behind global powerhouses like the U.S., Europe, and China when it comes to artificial intelligence (AI) advancements. However, with the introduction of Avataar’s new video AI model, Varya, there's hope for change.
To address this gap, the Indian government launched the India AI Mission, a $1.2 billion initiative aimed at supporting startups and making AI more accessible. One such beneficiary is Avataar AI, which has developed Varya to cater specifically to local contexts, including identifying regional festivals, food, and clothing.
Avataar didn't start from scratch; they took the publicly available Wan 2.2 model by Alibaba and refined it through a process called distillation, making Varya run faster and more cost-effectively. The result is a model that can generate a five-second 720p clip in just 45 seconds using an Nvidia H200 GPU—10 times faster than Wan 2.2's original 1,230 seconds. This speed and efficiency are game-changers for businesses and individuals alike.
More impressively, Avataar has priced Varya at ₹0.48 ($0.005) per second of video on their hosted service. Compare that to other models like Veo, Kling, Luma, and Runway, which typically charge $0.10 or more per second. That's a staggering 20x difference in cost.
According to Rajan Anandan, managing director of Peak XV: “India is a video-first market. We see this across every large consumer internet product in India: video wins over text. Current AI video models are too expensive for population-scale use in India. If video AI is going to reach students, teachers, MSMEs, creators, enterprises, and public services, costs have to come down dramatically.”
Avataar has also addressed the cultural nuances issue that often plagues image and video generation models. They've used curated data to train Varya to recognize specific Indian cultural elements like food, clothing, architecture, and festivals. This makes it a more relevant tool for local businesses and creators.
Varya will be released as an open-source model on India's AIKosh portal, allowing developers to self-host or modify the model according to their needs. Avataar is also making it available to its enterprise customers and exploring partnerships with video tools like Higgsfield and Adobe Firefly.
The launch of Varya marks a significant shift in India’s AI strategy, focusing on creating practical applications rather than competing at the foundational level. It's part of a broader government push to boost local innovation by providing cost-efficient compute resources and a robust developer ecosystem.
