India’s journey in artificial intelligence development has, until recently, progressed at a more measured pace compared to global leaders like the U.S., Europe, and China. While a handful of pioneering startups have emerged, their focus has largely been on large language models or voice technologies. To galvanize further innovation and bridge this gap, the Indian government launched the ambitious India AI Mission, a substantial $1.2 billion initiative designed to fuel the nation’s AI ecosystem.
A key component of this mission offers selected startups access to subsidized GPU compute resources, contingent on them releasing their models publicly. Among the twelve trailblazing companies chosen for this prestigious program is Avataar AI, a startup already making waves in e-commerce video tools. Avataar AI has now unveiled its groundbreaking new video model, appropriately named Varya, specifically engineered to resonate with local Indian contexts and scale.
Varya: Designed for India’s Diverse Tapestry
What makes Varya truly stand out is its deep understanding of India’s vibrant cultural landscape. Unlike many generic AI models, Varya is meticulously trained to recognize and integrate local nuances, from identifying distinct festivals and regional cuisines to understanding diverse clothing styles. This cultural awareness ensures that generated content is not only relevant but also authentic to the Indian user experience.
Avataar AI, backed by Peak XV, didn’t build Varya from the ground up, but rather leveraged existing innovation to create something uniquely powerful. They began with Alibaba’s publicly available video generation model, Wan 2.2, and employed a sophisticated technique called “distillation.” This process effectively compresses the original model’s capabilities into a leaner, faster, and more efficient version, perfectly optimized for Avataar’s specific applications.
Unlocking Unprecedented Speed and Affordability
The results of this innovative approach are nothing short of remarkable. Varya operates in just four steps, a dramatic reduction from Wan 2.2’s fifty steps, making it significantly more agile. This efficiency translates directly into performance gains: Varya generates video content ten times faster and at an astonishingly low fraction of the cost.
To put this into perspective, using an NVIDIA H200 GPU, Varya can generate a 5-second 720p clip in a mere 45 seconds. This pales in comparison to Wan 2.2, which would take an extensive 1,230 seconds for the same task. The speed and efficiency are game-changers for large-scale video production.
Perhaps the most compelling aspect of Varya is its projected pricing model. Avataar AI plans to charge an incredible ₹0.48 (approximately $0.005) per second of video on its hosted service. This makes Varya exponentially more affordable than leading international models like Veo, Kling, Luma, and Runway, which typically command $0.10 or more per second – representing a staggering 20x price difference.
As Rajan Anandan, managing director at Peak XV, succinctly put it, “India is a video-first market. Current AI video models are too expensive for population-scale use in India.” He emphasizes that drastically reducing costs is the biggest catalyst for AI adoption, enabling video AI to reach students, teachers, MSMEs, creators, enterprises, and public services across the nation.
Cultural Nuances and Open Access for All
A persistent challenge with global image and video generation models has been their tendency to overlook cultural nuances, often resulting in generic or even stereotypical outputs. Avataar AI directly addresses this by training Varya with meticulously curated data, enabling it to recognize intricate cultural details including local food, clothing, architecture, and festivals. This ensures content is not just generated, but truly resonates.
In line with the India AI Mission’s vision, Varya will be released as an open-weight model on India’s AI Kosh portal. This centralized government repository will also host Varya’s training data, empowering developers to self-host, modify, and innovate upon the model for their unique requirements. Avataar AI is also making Varya accessible to its enterprise clients and is actively exploring collaborations with prominent video tools such as Higgsfield and Adobe Firefly.
Anyone interested can already experience Varya firsthand on Avataar AI’s website, utilizing text prompts or reference images to generate their own custom video content. This open approach not only democratizes access to advanced AI capabilities but also fosters a collaborative environment for further innovation within India’s developer ecosystem.
India’s Grand AI Vision Takes Shape
The launch of Varya beautifully encapsulates a core philosophy guiding India’s AI aspirations: prioritizing the creation of impactful applications and a robust developer ecosystem over direct competition in foundational model development. This pragmatic approach acknowledges historical challenges such as limited compute infrastructure and data availability, focusing instead on where India can truly excel.
The India AI Mission is a crucial pillar in the government’s broader strategy to close this technological gap. Last year’s selection of twelve innovative startups, including Avataar AI, for cost-efficient compute access underscores this commitment. Furthermore, IT minister Ashwini Vaishnaw recently articulated India’s ambitious goal to attract $200 billion in AI investment by 2028 and more than double its GPU capacity within the next six months. With initiatives like Varya, India is well on its way to cementing its place as a formidable force in the global AI landscape.
Source: TechCrunch – AI
