Avataar AI has launched Varya, a new video generation model designed to make AI video cheaper, faster, and more culturally aware for India’s mass market.
The Peak XV-backed startup is one of 12 companies selected under the India AI Mission, a government-backed initiative that gives chosen startups subsidized GPU access in exchange for publicly releasing their models. Avataar is using that support to take on one of the biggest barriers in AI video: cost.
Varya is not meant to compete with the biggest global video models only on cinematic quality. Its pitch is different. It is built for speed, affordability, local context, and broad adoption across a market where video dominates consumer internet behavior.
That focus matters because India is one of the world’s largest video-first digital markets. Short videos, product demos, creator content, education clips, regional language content, social commerce, and mobile-first advertising all depend heavily on visual media. If AI video remains expensive, it may stay limited to large brands and well-funded creators. Avataar wants to bring the cost low enough for small businesses, educators, creators, students, enterprises, and public-service use cases.
Avataar did not build Varya completely from scratch. The company started with Wan 2.2, a publicly available video generation model from Alibaba, and used distillation to create a smaller and faster version optimized for Avataar’s use cases.
Distillation is a technique where the capabilities of a larger model are compressed into a leaner system. The goal is to preserve enough quality while making the model cheaper and faster to run.
That is the core of Varya’s advantage. Wan 2.2 normally runs through 50 steps, while Varya runs in four steps. Avataar says this makes the model about 10 times faster while reducing the cost sharply.
The performance difference is significant. On an Nvidia H200 GPU, Varya can generate a five-second 720p video clip in 45 seconds. Wan 2.2 takes about 1,230 seconds for the same kind of output.
That speed gap is central to the product strategy. In markets where creators and businesses need many short videos quickly, generation time matters almost as much as output quality. A tool that takes minutes instead of more than 20 minutes per clip can change how often people are willing to use it.
Varya’s biggest claim is price.
Avataar plans to charge ₹0.48 per second of video on its hosted service, which is roughly half a U.S. cent. That is far below many leading video generation tools, which often charge around $0.10 or more per second.
That difference could make AI video more practical in India. A five-second clip at Varya’s planned hosted price would cost only a few rupees. For small sellers, teachers, student creators, local advertisers, and regional content teams, that kind of pricing can make experimentation realistic.
This is important because AI video is still expensive compared with text or image generation. Video models require heavy compute because they must generate motion, consistency, frames, style, and sometimes object behavior across time. That cost has limited how widely video AI can be used, especially outside premium creative and enterprise settings.
Avataar’s argument is that video AI will only reach population scale in India if the price falls dramatically. The company is not only selling a model. It is trying to build an economic case for AI video in a country where the user base is huge but willingness to pay per clip can be much lower than in the U.S. or Europe.
Avataar is also positioning Varya around cultural understanding.
Many global AI image and video models can struggle with local details. They may produce generic Indian scenes, flatten cultural differences, miss regional clothing styles, misunderstand festivals, or rely on stereotypes. For a country as diverse as India, that is a serious limitation.
Avataar says Varya has been trained with curated data to better recognize Indian cultural context, including food, clothing, architecture, and festivals. That could make the model more useful for local advertising, regional storytelling, education, e-commerce, and creator content.
This is more than a cosmetic improvement. For AI video to work in India, it must understand what users actually want to show. A small business making a festival promotion, a teacher creating a local-language explainer, or a creator producing regional content needs outputs that feel familiar, not imported.
Cultural accuracy can become a product advantage. If a model understands local settings better than global competitors, it may be more useful even if it is not the most advanced video model in the world.
Avataar already focuses on AI video tools for e-commerce, and Varya fits naturally into that business.
Product videos are one of the clearest use cases for cheaper video generation. Online sellers often need many variants of product visuals, including different backgrounds, use cases, angles, styles, campaigns, and seasonal promotions. Producing those videos manually can be slow and expensive.
AI-generated product video can help sellers create more content without hiring production teams for every asset. A clothing seller could generate festival-themed product clips. A furniture brand could show a product in different home settings. A beauty brand could create short ad variations. A marketplace seller could turn basic product material into social-ready video.
For India’s small and medium businesses, this could be especially useful. Many sellers operate with limited marketing budgets but need to compete in visually crowded online channels. Cheaper video AI could help them produce more assets for commerce, ads, WhatsApp marketing, and social platforms.
That may be where Varya finds its first serious market. Consumer curiosity may bring attention, but e-commerce and advertising workflows can create repeat usage.
Avataar plans to release Varya as an open-weight model on India’s AI Kosh portal, along with training data. That means developers will be able to self-host, modify, and build on the model instead of only using Avataar’s hosted service.
This is important for India’s AI ecosystem. The country has often been described as strong in software services and applications but slower in foundational AI model development compared with the U.S., China, and Europe. Open-weight releases can help developers experiment, adapt models for specific industries, and build products without starting from zero.
The AI Kosh release also ties Varya directly to the India AI Mission’s broader goal. The government wants subsidized compute to produce public AI assets that can support startups, researchers, and developers across the country.
If developers adopt Varya, the model could become part of a wider ecosystem of video tools, local-language content apps, commerce platforms, education products, and creative workflows. That would make the project more meaningful than one startup product launch.
Varya’s launch also reflects a broader reality about India’s AI ambitions.
India may not immediately compete with the largest U.S. or Chinese labs on the scale of frontier foundation models. Compute constraints, capital requirements, talent concentration, and high-quality training data remain real challenges. But India can compete by building practical, lower-cost, locally optimized AI systems for huge real-world markets.
That is what Varya represents. It is not trying to be the most powerful video model in every category. It is trying to solve a specific local problem: AI video is too expensive and not culturally aware enough for India-scale use.
This may be a more realistic path for many Indian AI startups. Instead of chasing the biggest global model race, they can focus on affordability, language coverage, local behavior, cultural context, mobile-first use, and industry-specific applications.
That approach could still produce major companies. India’s internet market is large enough that a model optimized for local use can become commercially important without needing to beat every global model on benchmarks.
Varya enters a competitive AI video market.
Global companies such as Google, Runway, Luma, Kling, and others are pushing quickly into video generation. These companies have deep resources, strong research teams, advanced models, and growing developer ecosystems. Some are focused on cinematic quality. Others are building creative tools, world models, and enterprise-grade media workflows.
Avataar’s challenge is to avoid being squeezed between global model quality and local pricing pressure. If global models become cheaper quickly, Varya’s cost advantage may narrow. If Varya’s output quality is too limited, users may still choose more expensive tools for higher-stakes projects.
The company’s best path is likely specialization. If Varya can be fast, cheap, culturally useful, and easy to integrate into Indian commerce and creator workflows, it does not need to beat every global video model. It needs to be the most practical choice for the customers it targets.
That is a different kind of AI competition. The winner is not always the model with the best demo. It can be the model with the best fit for the market.
Avataar’s Varya launch shows how AI video may evolve over the next few years.
The first phase of AI video was about proving that models could generate impressive clips. The next phase is about making those clips affordable, fast, controllable, and locally relevant enough for real use.
That is especially true in markets like India, where scale can be enormous but pricing must be realistic. A model that costs too much per second will struggle to reach students, small businesses, regional creators, educators, and public-service teams. A model that ignores cultural nuance will struggle to produce content people actually want to use.
Varya is Avataar’s answer to both problems. It is cheaper, faster, and trained with local context in mind.
The model’s success will depend on whether users find the output good enough for everyday production and whether developers build on the open-weight release. But the direction is clear. India’s AI market may not only be about building bigger models. It may be about building models that fit India’s cost, culture, and scale.
For Avataar, that could be the real opportunity. If AI video becomes a mass-market tool in India, the winning model may not be the one that looks most impressive in a global demo. It may be the one people can actually afford to use every day.
Share your thoughts about this article.
Be the first to post a comment!
