Avataar introduces affordable and fast Varya video AI model for India

Indian AI startup Avataar has announced Varya, a new video generation model designed to understand local culture and operate at high speeds. This model is one of 12 projects selected under the Indian government's $1.2 billion India AI Mission program. Varya was created using distillation techniques based on the open-source Wan 2.2 model from Alibaba, making it 10 times faster than the original. This is reported by Techcrunch.com reports .

Technically, Varya can generate a 5-second 720p video on an NVIDIA H200 GPU in just 45 seconds. For comparison, the Wan 2.2 model requires 1230 seconds for the same task. Most importantly, the model's cost is significantly lower than competitors. Avataar charges $0.005 per second of video, which is nearly 20 times cheaper than global platforms like Veo, Kling, Luma, and Runway.

The main advantage of the Varya model is its adaptation to the Indian cultural environment. It accurately recognizes local festivals, national dishes, clothing, and architectural styles. While many global models often rely on stereotypes and make mistakes, Avataar has solved this issue using specially curated datasets. This opens up great opportunities for e-commerce and content creators.

Currently, the Varya model is hosted as an open-weights model on India's AI Kosh portal. This means developers can deploy it on their own servers or modify it according to their needs. Avataar also plans to integrate this technology with tools like Higgsfield and Adobe Firefly. Users can test the model on the company's website using text prompts or reference images.