Avataar AI Disrupts Video Generation with Culturally Aware Model
Avataar AI has launched Varya, a new video generation model designed for the Indian market. By using a technique called distillation, the company made the model ten times faster and significantly cheaper than existing competitors. Unlike global AI models, Varya is trained to recognize Indian cultural details like festivals and clothing. The model is part of a government initiative to boost AI development and is available for developers to use and modify.
India’s AI model output has been relatively slow compared to global rivals like the U.S. and China. To address this gap, the government introduced the India AI Mission, which provides selected startups with subsidized GPU compute. Avataar AI, a startup backed by Peak XV, was recently selected to participate in this initiative. The company has now launched a video generation model called Varya, which is specifically optimized for India's population-scale needs.
Instead of building a model from scratch, Avataar AI utilized Alibaba’s Wan 2.2 as a foundation. By using a technique called distillation, they compressed the model’s capabilities into a leaner version. The result is a model that is optimized for specific use cases and runs significantly faster. While Wan 2.2 requires 50 steps to produce a video, Varya only needs four. Consequently, Varya generates content 10 times faster and at a fraction of the original cost. On an NVIDIA H200 GPU, a 5-second 720p clip can be generated in just 45 seconds.
One of the most impressive features of Varya is its cultural awareness. Generic models often produce stereotyped or generic outputs regarding architecture and clothing. However, Avataar AI used curated data to ensure Varya recognizes local nuances, such as Indian festivals and food. Rajan Anandan, managing director at Peak XV, stated that current AI video models are too expensive for India. He argued that bringing down costs is the biggest unlock for AI adoption among students and enterprises.
Varya will be released as an open-weight model on India’s AI Kosh portal. This allows developers to modify the model for their own requirements. Furthermore, the government aims to attract $200 billion in AI investment by 2028. Although model development has been slow due to a lack of compute, these new initiatives are expected to close the gap. Anyone can currently test Varya on the Avataar AI website using text prompts.
Take a position. Out loud, if you can.
Four ways to start. Pick one and try saying it before you scroll on.
Tip · Record yourself, use in a notebook, or practice with a language partner.
How does Varya's efficiency compare to the Wan 2.2 model?
Passive Voice
Used when the action is more important than the person performing it.
“India’s AI model output has been relatively slow compared to global rivals like the U.”
What to know · B2
Try saying this aloud
Scenario: Critically evaluating a new product.
- 01“It is optimized for specific use cases.”
- 02“This addresses a gap in the market.”
- 03“It runs significantly faster.”
Register tip · formal
🔑Key Phrases
To fix a problem or missing part.
The new law seeks to address this gap in safety.
Made better for a particular purpose.
The software is specifically optimized for tablets.
High-quality, selected information.
We used curated data for the research project.
🎙️ Article Audio — Kokoro TTS
Avataar AI Disrupts Video Generation with Culturally Aware Model
Adapted from TechCrunch · Read the original. LectoPress rewrites the facts as original graded-reader text for language learners.
Get stories at your level, every day
B2 · EN · delivered to your inbox · unsubscribe any time
Customize language, level & topics → full preferences
