New Delhi: Avataar.ai on Friday unveiled Varya, an indigenous video-generation AI model developed with support from the IndiaAI Mission. The company claims Varya can produce videos at a fraction of the cost of several leading global models.
Technical Innovation
Varya employs a machine-learning technique called distillation to reduce the number of computational steps required for video generation from more than 50 to just four. This approach lowers computing costs while maintaining similar output quality. The model was launched in New Delhi in the presence of MeitY secretary S Krishnan.
Cost Efficiency
Avataar stated that Varya can generate video at approximately ₹0.48 per second. Based on internal benchmarks, it is up to ten times more cost-efficient than several international models. The company noted that many recent AI video-generation systems from other tech firms involve high computing costs.
Distillation Process
Distillation allows a smaller AI model to achieve results comparable to a much larger model while using significantly less computing power. Conventional video-generation systems typically require over 50 iterative steps, whereas Varya completes the process in just four steps.
Indian Context Training
The model was trained to generate visuals reflecting Indian settings, including festivals, clothing, food, public spaces, and regional contexts. Avataar highlighted potential applications for teachers creating visual lessons, small businesses developing advertisements, and government agencies communicating information through videos.
Government Support
S Krishnan remarked, "The launch of one of the foundational models supported under the IndiaAI Mission marks a significant milestone in India's AI journey." He emphasized the government's commitment to building indigenous AI capabilities and fostering innovation through foundational AI models.
Avataar CEO and co-founder Sravanth Aluru stated, "For a country of 1.4 billion people, affordability is not a feature, it is a prerequisite." He noted that India's AI opportunity would be shaped not only by the largest models but also by those that can be used widely at lower cost.
User Features
Users can enter a text prompt or upload an image to generate a video and extend it through additional clips. The technology is applicable to lessons, advertisements, informational content, and other forms of visual storytelling. Avataar plans to publish a technical report detailing Varya's architecture, distillation methodology, and performance benchmarks.
IndiaAI Mission
Varya is among the first AI foundation models announced under the IndiaAI Mission support programme, which aims to encourage the development of indigenous AI technologies and applications tailored to Indian requirements. Avataar was selected under the mission to develop indigenous AI capabilities using subsidised national computing infrastructure.
