Avataar AI launches Varya, a video generation model built for India's market that runs 10 times faster and costs roughly 20 times less than global competitors.

What happened: Avataar AI, one of 12 startups selected for India's government-backed AI Mission, released Varya, a video model built by compressing Alibaba's Wan 2.2 model using a technique called distillation. Varya generates a 5-second 720p video in 45 seconds using an NVIDIA H200 GPU, versus 1,230 seconds for Wan 2.2, and the company plans to charge ₹0.48 ($0.005) per second of video—far cheaper than competitors like Veo, Kling, Luma, and Runway, which typically charge $0.10 or more per second.

Why it matters: Video AI tools have been too expensive for broad adoption in India, where video dominates over text in consumer internet products. Avataar trained Varya on curated data to recognize cultural nuances—food, clothing, architecture, and festivals—addressing a gap where existing models often produce stereotyped or generic outputs. The dramatic cost reduction appears intended to unlock video AI use among students, teachers, creators, and small businesses across India's population scale.

What to watch: Varya will be released as an open-weight model on India's AI Kosh portal (the government's centralized repository for AI models and datasets) along with its training data, so developers can self-host or modify it. The model is already available to try on Avataar's website using text prompts or reference images, and Avataar plans to make it available to enterprise customers and is open to partnerships with tools including Higgsfield and Adobe Firefly.