Varya AI Model: India’s First Distilled Video-Story Generator
Why in the news
The IT ministry presented a homegrown AI tool that turns text or images into videos. It aims to make advanced generative AI affordable by lowering the barriers of cost, language and hardware.
Key facts
- Model: Varya, a generative text-to-video and image-to-video platform.
- Builder: Avataar; unveiled by: Ministry of Electronics and Information Technology (MeitY).
- Speed: 4 steps instead of 50 through distillation.
- Cost: about ₹0.48 per second of video.
| Feature | What it means |
|---|---|
| Distilled generation | A large model is compressed into a lighter one, improving speed and efficiency |
| Cost efficiency | Sharply lower content-creation cost |
| Cultural awareness | Accurate depiction of regional festivals, clothing, food, languages and daily life |
| Idea to Video to Story | Create clips from text or images, then extend with more prompts into long-form narratives |
Related concepts
- Model distillation: a smaller student model learns to mimic a larger teacher model, giving faster inference and lower memory and compute cost.
- IndiaAI Mission: approved in March 2024 with ₹10,372 crore, implemented by MeitY; components include compute capacity, datasets platform, FutureSkills, startup financing and safe and trusted AI. AIRAWAT is the national AI supercomputer.
- Bhashini: National Language Translation Mission under MeitY, a digital public infrastructure for Indian languages covering 22 scheduled languages, launched in 2022.
Exam angle
- Developer: Avataar; ministry: MeitY.
- Key technical term: distillation (4 steps vs 50).
- Cost per second of video: about ₹0.48.