Key facts
- Avataar AI has launched Varya, a new video generation model tailored for the Indian market.
- Varya is significantly faster and cheaper than existing AI video models, costing approximately $0.005 per second.
- The model was developed by distilling Alibaba's Wan 2.2 model, making it more efficient.
- Varya is trained to recognize Indian cultural nuances, including festivals, food, and clothing.
- The model will be released as an open-weight model on India's AIKosh portal, allowing developers to modify it.
Avataar AI, an Indian startup backed by Peak XV, has launched Varya, a new video generation model designed to be culturally aware and cost-effective for the Indian market. The model is built using a distillation technique on Alibaba's publicly available Wan 2.2 model, resulting in a significantly faster and cheaper product.
Varya can generate video in four steps, making it 10 times faster than its predecessor and costing approximately $0.005 per second, a stark contrast to competitors charging $0.10 or more. This focus on affordability is crucial for widespread AI adoption in India, a video-first market.
The model has been trained on curated data to recognize Indian cultural nuances, including festivals, food, and clothing, addressing a common issue of generic or stereotyped outputs from AI video generators. Varya will be released as an open-weight model on India's AIKosh portal, allowing developers to self-host or modify it.
Avataar AI's approach aligns with India's broader AI strategy, which emphasizes application development and ecosystem building rather than competing directly on foundation models due to limitations in compute and data availability. The India AI Mission, a $1.2 billion initiative, aims to support startups like Avataar AI by providing subsidized GPU compute.