Indian startup Avataar AI launched Varya in June 2026, a video generation model designed to produce culturally aware content at a fraction of the cost of competing services. The company plans to charge ₹0.48 ($0.005) per second of video — roughly 20 times cheaper than models such as Veo, Kling, Luma, and Runway, which typically charge $0.10 or more per second.
Avataar AI, backed by Peak XV and focused on video tools for e-commerce, did not build Varya from scratch. The company started with Wan 2.2, a publicly available video generation model released by Alibaba, and applied a technique called distillation — compressing the model’s capabilities into a leaner, faster version. The result runs in four steps rather than Wan 2.2’s 50, generating video 10 times faster. Using an NVIDIA H200 GPU, Varya produces a 5-second 720p clip in 45 seconds, compared to 1,230 seconds for Wan 2.2.
Beyond speed and cost, Avataar says Varya was trained on curated data to recognize Indian cultural nuances including food, clothing, architecture, and festivals — addressing a widely noted limitation of general-purpose image and video models.
Avataar AI is one of 12 startups selected for India’s AI Mission, a roughly $1.2 billion government initiative that provides selected companies with subsidized GPU compute in exchange for publicly releasing their models. Varya will be released as an open-weight model on India’s AI Kosh portal, along with its training data, allowing developers to self-host or modify it. Avataar also plans to offer the model to enterprise customers and says it is open to partnerships with tools including Higgsfield and Adobe Firefly.
“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,” said Rajan Anandan, managing director at Peak XV. The launch comes as India’s government has stated a goal of attracting $200 billion in AI investment by 2028 and more than doubling GPU capacity within six months.
Source: TechCrunch