NextFin News - Avataar AI on June 11 unveiled Varya, a video-generation model it says is designed for India’s scale, with hosted pricing set at ₹0.48, or about $0.005, per second of video, according to TechCrunch.
That is roughly 20 times cheaper than the $0.10-a-second pricing often associated with models such as Veo, Kling, Luma and Runway. Avataar is making that case in a market it describes as one where video, not text, is the default medium for consumer attention.
Rajan Anandan, managing director at Peak XV, told TechCrunch that India is a “video-first market” and that current AI video models are too expensive for population-scale adoption across students, teachers, MSMEs, creators, enterprises and public services. For Avataar, the pricing is less a technical flourish than a commercial argument: low enough, it hopes, to make video generation usable at scale.
Varya was not built from scratch. Avataar started with Wan 2.2, a publicly available video-generation model released by Alibaba, and used distillation to compress its capabilities into a leaner system tailored to the company’s use cases. Avataar says Varya runs in four steps rather than Wan 2.2’s 50. It claims that makes output about 10 times faster and materially cuts compute costs.
The company’s performance numbers are central to that pitch. Using an NVIDIA H200 GPU, Varya can generate a five-second 720p clip in 45 seconds, versus 1,230 seconds for Wan 2.2, according to Avataar’s figures cited by TechCrunch. For customers making ad creative, testing concepts or producing localized content, that changes how often they can generate and revise video. In India, where many buyers are unlikely to pay enterprise-grade Western model rates for routine video work, price and speed are part of the product.
Avataar is also arguing that Varya is better tuned to Indian visual context than broader global models. The company says it curated data to teach the model cultural nuances across food, clothing, architecture and festivals. That matters in a country where visual identity varies sharply by region, religion and language. A system that can render a sari, a wedding stage or a neighborhood storefront without slipping into generic imagery or caricature may be more useful to advertisers and merchants than one built mainly to perform well in benchmark demos.
That is especially relevant in e-commerce, where Avataar already has experience. The startup has focused on video tools for online retail, where product visualization can affect conversion rates and localized content has direct monetization value. For merchants, the appeal is straightforward: faster creative generation, lower production costs and assets that fit campaigns aimed at Indian cities, tier-two markets and vernacular audiences. If Varya can shorten turnaround times while keeping spending low, it may appeal to sellers that cannot justify custom studio work for every product variant or promotion.
The opening is clear, but so are the limits. A low per-second price does not guarantee a durable business if customers use the model only occasionally or treat it as a novelty instead of folding it into regular workflows. Avataar still needs recurring demand from brands, agencies and platforms to turn technical efficiency into revenue. Its India-specific positioning could help distinguish the product, but it could also narrow the market if the best economics depend on buyers that are highly price-conscious and spread across a fragmented landscape.
Competition is another test. The global video-AI market already includes better-funded rivals, and several of the large model names Avataar compares itself with have scale, brand recognition and international distribution. Avataar’s advantage, if it lasts, lies in specialization: a model tuned to India’s linguistic and visual context and priced for volume instead of premium use. Distillation can make a model efficient. It does not ensure competitors cannot copy the same method or cut prices later.
Avataar is effectively making a case for a different AI adoption path in India. Rather than asking customers to pay Silicon Valley prices for generalized video generation, it is trying to push costs down until video becomes routine infrastructure for commerce, education and public-facing services. Whether that holds up beyond a demo will depend on the same numbers Avataar is emphasizing now: a four-step architecture, a 45-second output time on an H200 GPU and training aimed at India-specific use cases.
Explore more exclusive insights at nextfin.ai.
