Startup Avataar.ai has launched Varya AI, India's first distilled video AI model developed under the IndiaAI Mission. According to reports from Business Standard, the model was launched in the presence of S Krishnan, secretary at the Ministry of Electronics and Information Technology, who stated that the model represents a significant milestone in the country's efforts to build indigenous foundational AI models. Founder Sravanth Alluru emphasized that contemporary leading video models are expensive to run and often under-represent the cultural, linguistic and visual diversity of India, with Varya built to address both cost and context challenges.
As reported by AI Chat Daily, Varya AI generates 5-second 720p clips in 45 seconds on an NVIDIA H200 GPU, representing a 27x speed improvement over Alibaba's Wan 2.2 model. According to Business Standard, the model can generate up to 211 seconds of video for every ₹100 spent, which is far higher than existing open-source and proprietary alternatives. The company has priced the hosted service at ₹0.48 ($0.005) per second of video, which is approximately 20x cheaper than competitors like Veo, Kling, Luma, and Runway, who typically charge $0.10 or more per second. This pricing strategy addresses what Peak XV managing director Rajan Anandan calls a structural problem for India's market, where consumer products skew video-first but current frontier video models price out population-scale use cases.
According to AI Chat Daily, Varya achieves its performance improvements through compression rather than new architecture. The startup started with Alibaba's Wan 2.2 model and applied distillation to collapse the inference path from 50 steps down to 4 steps, generating video approximately 10 times faster than the parent model while targeting Avataar's specific use cases in e-commerce content. The model will also be released as an open-weight model on India AI Kosh, the government's centralized repository for publicly available AI models, allowing developers to self-host or fine-tune it without going through Avataar's API. As per Business Standard, the startup will publish a technical report outlining Varya's model architecture, distillation methodology and benchmarks.
According to Business Standard, Avataar.ai is among 12 startups selected for the IndiaAI Mission, a government program with a total outlay of ₹10,300 crore over five years launched last year to bolster the country's leadership in AI and democratise the technology's benefits. The program offers subsidized national AI compute infrastructure, with Avataar gaining access to this infrastructure as part of the second phase selection. This regulatory compliance positions the model for broader adoption in the Indian market, with the company building regulatory safeguards including restrictions on generating NSFW content, watermarking of AI-generated videos, and content labelling features aligned with emerging government requirements on synthetic media.
As reported by Business Standard, Varya has been trained to better understand Indian contexts, including local clothing, facial features, festivals, streetscapes, and cultural settings. The company showcased examples highlighting the model's ability to generate more India-specific visual outputs than global video-generation models. According to Business Standard, CEO Sravanth Alluru stated that the model will make high-quality AI video creation more cost-efficient for education, creators, MSMEs (micro, small, and medium enterprises), and the country's overall digital economy. For the global video-generation market, this credible open-weight option priced 20x below incumbents reshapes the floor of what enterprise buyers will pay per second, particularly in price-sensitive regions outside the U.S. and Europe. The broader policy backdrop includes IT minister Ashwini Vaishnaw's goal of attracting $200 billion in AI investment by 2028 and more than doubling India's GPU capacity within six months.
