Avataar AI launches low-cost, culturally aware video model Varya in India

Avataar AI, a startup selected for the Indian government’s $1.2 billion India AI Mission, has launched Varya, a video generation model optimised for the Indian market. Backed by Peak XV, the company has positioned the model as a solution to the high costs and cultural disconnects that currently limit AI video adoption in the region. Varya is designed to recognise local nuances, including specific festivals, food, clothing, and architecture, addressing a common failure point for global models that often produce stereotyped or generic outputs.

The model was not built from scratch but created through a distillation process applied to Alibaba’s open-source Wan 2.2 model. This technique compressed the original model’s capabilities into a leaner version that runs in four steps, compared to Wan 2.2’s 50 steps. Benchmarks using an NVIDIA H200 GPU demonstrate the efficiency gains: Varya can generate a 5-second 720p clip in 45 seconds, whereas the original Wan 2.2 required 1,230 seconds. This results in a model that is approximately 10 times faster than its predecessor.

Pricing is a central feature of Varya’s market entry. Avataar AI plans to charge ₹0.48 ($0.005) per second of generated video on its hosted service. This rate is significantly lower than major global competitors such as Veo, Kling, Luma, and Runway, which typically charge $0.10 or more per second. Rajan Anandan, managing director at Peak XV, noted that cost is the primary barrier to AI adoption in India, stating that video AI must become affordable for students, teachers, micro-enterprises, and public services to achieve population-scale use.

The model will be released as an open-weight model on the government’s AI Kosh portal, alongside its training data. This move allows developers to self-host or modify the model for their own needs, aligning with the India AI Mission’s goal of fostering a robust domestic developer ecosystem. Avataar AI is also making the model available to enterprise customers and has expressed openness to partnerships with video tools including Higgsfield and Adobe Firefly.

Varya’s launch reflects a strategic pivot in India’s AI ambitions, focusing on application development and ecosystem building rather than competing directly on foundation models. The India AI Mission, which selected 12 startups for subsidized GPU compute, aims to close the gap in compute and data availability. This initiative supports broader government targets, including attracting $200 billion in AI investment by 2028 and doubling GPU capacity within six months, as outlined by IT minister Ashwini Vaishnaw.

The model is currently available for public testing on Avataar’s website via text prompts or reference images. By providing an open-weight version on AI Kosh, Avataar AI aims to lower the barrier to entry for Indian developers, enabling them to build culturally relevant applications without the prohibitive costs associated with proprietary global models.