Varya: AI model for video is 10x faster and made for India

The Indian startup Avataar launched the Varya, an artificial intelligence model for video generation that promises to be ten times faster than the Alibaba Wan 2.2 and at the same time take into account local cultural aspects. The model was one of the 12 selected by the India AI Mission program, which has an investment of $1.2 billion from the Indian government.

Why is speed crucial in AI models for video?

In a scenario where demand for video content grows exponentially, latency is one of the biggest bottlenecks. Traditional models such as Wan 2.2 can take minutes to generate video seconds, making applications impossible in real time or at low cost. Varya reduces this time in an order of magnitude, making video generation via AI more accessible to small businesses and individual creators. As we discussed in The era of 'all you can consume' of AI is over: it is time to count calories, computational efficiency becomes a crucial competitive differential.

How is Varya culturally conscious and relevant to India?

Unlike generic models predominantly trained in Western data, Varya was developed with a set of data that reflects India's linguistic and visual diversity. This means that it understands better traditional clothes, gestures, expressions and even regional nuances that global models often ignore. According to Avataar, this cultural adaptation reduces generation errors and makes content more relevant to the Indian public.

What does India AI Mission mean for the IA Varya model?

The India AI Mission program, with a contribution of $1.2 billion, aims to accelerate the development and adoption of AI in the country, financing projects that solve local problems. Varya's selection of only 12 initiatives brings credibility and resources to Avataar. The company plans to use support to scale the infrastructure and expand the model to other Indian languages and dialects, as well as making it available via API to developers.

Frequently Asked Questions about the IA Varya model

Is Varya available for public use?

Yes, Avataar plans to offer the model through its own API and platform, focusing on emerging markets.

What are the main differences between Varya and Wan 2.2?

Varya is ten times faster and has been trained with Indian cultural data, making it more suitable to generate videos that reflect local diversity.

How does Avataar guarantee the quality of the videos generated?

The company uses latency compression techniques and model optimization, combined with a set of cured data to avoid biases.

Source: TechCrunch