India’s First Homegrown AI Video Model ‘Varya’ Debuts: All You Need to Know

Avtaar.ai developed India's first indigenously ‘Varya’  to video generation time from 50 steps to 4 making it ten times efficient over the mains models at affordable price. It has been developed under the India AI Mission. 

Varya AI tool can be used in Education to E-commerce sectors by bringing AI video generation services to India at large scale. It has immense potential to convert text into videos by providing smart prompting. 

Standard global video diffusion models are computationally heavy, often iterating through more than 50 complex, noisy steps before rendering a clean and final video frame. This intensive process makes video generation sluggish and prohibitively expensive for individual creators and small businesses.

Avataar.ai solved this infrastructure bottleneck using a cutting-edge machine learning technique through the model distillation. Avataar took a foundational public architecture (Alibaba's Wan 2.2) and distilled its capabilities into a lean, ultra-efficient student model. 

Varya reduces the required generation process from 50 steps down to just 4 steps. The model achieves comparable frontier-quality outputs while executing up to 10 times faster and consuming a fraction of the computing power. 

Unmatched affordability of ₹0.48 Per Second

It must be a core prerequisite for an AI model to succeed in a country of 1.4 billion people, affordability cannot merely be a feature while international alternatives like OpenAI’s Veo, Kling, Luma or Runway typically charge $0.10 (₹8 to ₹10) or more per second of video, Varya completely disrupts the market dynamics. According to internal inference-cost benchmarks backed by Avataar.ai and MeitY:

Varya addresses a glaring limitation of Western and East Asian AI models like cultural misrepresentation. Global models trained heavily on Western datasets frequently struggle with local nuances often outputting stereotypical or inaccurate representations of Indian life.

Varya has been explicitly trained on curated datasets including over 40,000 Indian cultural data points to naturally recognize and accurately render local clothing styles like sarees, dhotis, regional attire, authentic Indian food and culinary details Regional architectural nuances and public settings and diverse festivals, communities and everyday streetscapes.

Varya operates on a seamless and  user-friendly prompt framework like Idea → Video → Story. Users can input simple text prompts or upload a base image to generate an initial video clip and then continuously extend the sequence by chaining additional clips together.

This smart prompting layout is designed to spark a massive transformation across critical Indian sectors education, e-commerce and MSME’s and other public services.