In News:

The Ministry of Electronics and Information Technology (MeitY) has unveiled "Varya", India's first indigenous distilled video story-generating Artificial Intelligence (AI) model, aimed at making advanced generative AI affordable, efficient, and culturally relevant.

What is Varya?

Varya is an indigenous foundation AI model designed to generate high-quality videos from text prompts (text-to-video) and images (image-to-video). Developed by Avataar, an AI-native transformation company, the model uses AI model distillation to deliver faster and cost-effective video generation while preserving high-quality outputs.

The platform enables users to transform simple ideas, written prompts, or uploaded images into engaging digital stories, thereby democratizing access to advanced generative AI technologies.

Objective

Varya aims to make high-end generative AI accessible across India by significantly reducing computational costs, hardware requirements, and language barriers. It seeks to promote inclusive AI adoption among educators, students, entrepreneurs, MSMEs, and content creators.

Key Features

A major innovation of Varya is its use of AI model distillation, a technique that compresses a large AI model into a smaller and more efficient version without substantial loss of performance. As a result, the model generates high-quality videos in only four inference steps, compared to nearly fifty steps required by conventional models.

The platform offers:

• Text-to-video and image-to-video generation.
• Low-cost video creation at approximately ?0.48 per second.
• Ability to extend short videos into long-form narratives through an "Idea → Video → Story" workflow.
• Strong Indian cultural contextualization, accurately depicting regional festivals, attire, food habits, languages, and everyday life.

AI Model Distillation

Model Distillation is a machine learning technique in which a large, complex AI model (teacher model) transfers its knowledge to a smaller, lightweight model (student model). The distilled model retains much of the original model's performance while requiring significantly lower computational resources, making deployment faster, cheaper, and more energy-efficient.