Avataar AI has launched Varya, a video AI model tailored for India’s cultural context. The model is designed to recognize local nuances such as festivals, food, and clothing. It is built on Alibaba’s Wan 2.2 using distillation techniques to create a faster, more cost-effective version. Varya runs in four steps instead of Wan 2.2’s 50, achieving a 10x speed boost. The model can generate a 5-second 720p clip in 45 seconds using an NVIDIA H200 GPU, compared to 1,230 seconds for Wan 2.2. Avataar AI plans to charge ₹0.48 ($0.005) per second for Varya, significantly lower than global models like Veo, Kling, Luma, and Runway, which typically charge $0.10 or more per second. This pricing strategy aims to make video AI accessible for a broader Indian audience, including students, teachers, and MSMEs. The model is also open-weight and will be released on India’s AI Kosh portal, allowing developers to self-host or modify it. Avataar AI is open to partnerships with video tools like Higgsfield and Adobe Firefly, and the model is already available for testing on its website. Varya’s launch reflects India’s strategy to focus on application development and a robust developer ecosystem rather than competing on foundation models. Industry experts note that India’s slower model development is due to limited compute and data availability, but the India AI Mission is helping to address these gaps. Last year, the government selected 12 startups, including Avataar AI, to develop AI models with subsidized compute. IT minister Ashwini Vaishnaw has also set a goal to attract $200 billion in AI investment by 2028 and increase GPU capacity within six months.
Source: techcrunch
