Most AI video tools are priced for Western ad budgets. A few seconds of footage from Google’s Veo, Kling or Runway can cost more than a small shop in India spends on marketing in a day. Varya is a direct challenge to that.
Built by Bengaluru startup Avataar under the government’s IndiaAI Mission and launched in New Delhi on 12 June 2026, it promises video generation at about Rs 0.48 (roughly half a US cent) per second, and visuals that actually understand Indian festivals, food, clothing and streets.
This review looks at what Varya is, how it works, what it really costs compared with global models, who it suits, and where it falls short. It draws on Avataar’s launch materials, the government’s announcement and independent reporting.
Quick Verdict
Varya is the cheapest serious AI video generator you can use today, and its India-focused training fills a real gap. It is a good fit for social clips, small business ads and classroom visuals in Indian settings. It is not a match for Veo, Kling or Seedance on motion realism, sound or long cinematic shots, and its quality and cultural accuracy claims still await independent testing.
What is Varya AI?
Varya AI is an Indian AI video-generation model developed by Avataar with support from the Government of India’s IndiaAI Mission. Launched on June 12, 2026, it converts text prompts and images into short video clips and lets users extend them into longer visual stories.
| Website | varya.avataar.ai |
| Developer | Avataar, a Bengaluru AI company backed by Peak XV Partners and Tiger Global |
| Launched | 12 June 2026, New Delhi, under the IndiaAI Mission |
| Model | 14-billion-parameter video model distilled from Alibaba’s open Wan 2.2 |
| Speed | 4 generation steps instead of 50; a 5-second 720p clip in about 45 seconds on one Nvidia H200 (company figure) |
| Output | 5-second clips from text or an image, extendable scene by scene |
| Price | About Rs 0.48 (US$0.005) per second, hosted |
| Open Weights | Announced for release on India’s AIKosh portal |
| Best For | Indian creators, small businesses, educators and enterprises needing low-cost video |
Varya is not built from scratch. According to TechCrunch, Avataar started from Alibaba’s publicly available Wan 2.2 video model and used distillation, training a model to reach similar results in far fewer steps. Standard diffusion video models refine each clip over about 50 steps. Varya does it in 4.
That single change is where the speed and price come from. Avataar says a five-second 720p clip that takes Wan 2.2 about 1,230 seconds on an H200 GPU takes Varya about 45 seconds, with comparable quality on its internal benchmarks. The company has said a technical report covering the architecture, distillation method and benchmarks will follow.
Avataar also trained the model on curated data about Indian life: regional clothing, food, architecture, festivals and public spaces. The aim is to avoid the stereotyped or generic results that global models often produce when asked for an Indian wedding, a Diwali market or a village classroom.
How Varya Works
The product follows a simple Idea, Video, Story flow:
- Sign in at varya.avataar.ai.
- Describe an idea in a text prompt, or upload a reference image such as a product photo.
- Generate a five-second clip.
- Extend the story by continuing from the last clip to add new scenes, building a longer sequence step by step.
This scene-by-scene approach suits short ads, reels and lesson visuals. Longer pieces need several continuations and some planning, much like storyboarding.
What Varya Costs Compared With Global Models
Using published per-second rates collected by Ngram, a minute of 720p video costs about $0.30 on Varya, $3 on Veo 3.1 Lite, $6 on Veo 3.1 Fast and $24 on Veo 3.1 Standard. Avataar’s own claims range from “up to 10x more cost-efficient” in the government release to “27 times cheaper than comparable open-source models” in founder interviews, so the exact multiple depends on what you compare against.
Table 2: What Common Video Jobs Cost on Varya
| Job | Length | Approx. Varya Cost | Approx. Veo 3.1 Fast Cost |
|---|---|---|---|
| Instagram reel or YouTube Short | 30 seconds | Rs 14 (about $0.15) | $3.00 |
| Product ad for a small shop | 15 seconds | Rs 7 (about $0.08) | $1.50 |
| Classroom explainer visuals | 2 minutes | Rs 58 (about $0.60) | $12.00 |
| Ten ad variations to test | 10 x 15 seconds | Rs 72 (about $0.75) | $15.00 |
Real costs run higher because you will regenerate clips that don’t work. Even at three or four attempts per usable clip, Varya stays far cheaper than the global options.
Where Varya Is Strong
- Price. At about half a US cent per second, experimenting is almost free, which changes how small teams can work.
- Indian context. Training on Indian clothing, food, festivals and places should mean fewer awkward or stereotyped results for local scenes.
- Speed. Four-step generation means clips in under a minute, according to Avataar.
- Open weights. A planned release on AIKosh would let developers self-host and fine-tune it, which global closed models don’t allow.
- Simple workflow. Text or image in, short clip out, then extend the story.
Where Varya Falls Short
- Not frontier quality. The Next Web notes that ByteDance’s Seedance, Kling and Alibaba’s own Wan push motion realism and audio generation well beyond what Varya offers.
- Short clips. Each generation is five seconds, so longer videos need many continuations.
- No native audio mentioned. Voiceover, music and sound effects need separate tools.
- Unverified claims. Quality parity, cultural accuracy and cost multiples come from Avataar’s internal benchmarks. The technical report and open weights were announced but not dated.
- Resolution and languages unclear. Public material focuses on 720p and doesn’t list supported prompt languages.
Varya vs Other AI Video Generators
Table 3: How Varya Compares
| Tool | Approx. Price per Second | Strength | Best For |
|---|---|---|---|
| Varya | $0.005 | Very low cost, Indian cultural context, planned open weights | Indian SMBs, creators, educators |
| Veo 3.1 (Google) | $0.05 to $0.40 | High realism and native audio | Premium ads and cinematic shots |
| Luma Ray3.2 | $0.06 | Fast, good camera motion | Creative and product shots |
| Kling | $0.10 or more | Strong motion and longer clips | Social content and storytelling |
| Wan 2.2 (self-hosted) | Your GPU cost | Open weights, full control | Developers with their own GPUs |
Who Should Use Varya
Table 4: Fit by User Type
| User | Good Fit? | Why |
|---|---|---|
| Small businesses in India | Yes | Product ads and reels for a few rupees each |
| Teachers and edtech teams | Yes | Visual lessons in familiar Indian settings |
| Indian creators and agencies | Yes, for drafts and social clips | Cheap testing of many ideas before polishing |
| Developers and startups | Yes, once weights are released | Self-hosting and fine-tuning on AIKosh |
| Global brands needing cinematic quality | Not yet | Veo, Kling or Seedance deliver more realism and audio |
| Long-form video producers | Partly | Five-second clips need a lot of stitching |
If you plan to publish Varya clips on YouTube, our YouTube SEO guide covers titles, descriptions and chapters, and these social media marketing statistics show where short video performs best.
Pros and Cons
Table 5: Varya Pros and Cons
| Pros | Cons |
|---|---|
| About Rs 0.48 per second, far below global rivals | Motion realism trails Veo, Kling and Seedance |
| Trained for Indian culture and settings | Five-second clips per generation |
| Clips in under a minute (company figure) | No audio generation mentioned |
| Open weights planned on AIKosh | Quality and cost claims not independently tested yet |
| Backed by the IndiaAI Mission and Peak XV | Technical report and weights release not dated |
How to Get Started
- Go to varya.avataar.ai and sign in.
- Start with a short, concrete prompt that names the setting, people, clothing and action, for example a tea stall in Kolkata at dawn during monsoon.
- Upload a product photo if you want the clip to feature your item.
- Generate several versions, keep the best, and extend it scene by scene.
- Add voiceover, music and captions in a separate editor before publishing.
FAQs
What Is Varya AI?
Varya is an AI video generation model from Bengaluru-based Avataar, launched in June 2026 under India’s IndiaAI Mission. It creates short videos from text or images and is trained to understand Indian cultural settings.
How Much Does Varya Cost?
Avataar prices hosted generation at about Rs 0.48, or roughly half a US cent, per second of video. A one-minute video costs about Rs 29 before retries.
Is Varya Free or Open Source?
Avataar has announced an open-weight release on India’s AIKosh portal, which would let developers download and self-host it. A release date had not been announced at launch.
Is Varya Better Than Google Veo?
It is far cheaper and more tuned to Indian settings, but Veo and other leading models deliver more realistic motion and native audio. Varya suits budget and volume; Veo suits premium quality.
Who Made Varya?
Avataar, a Bengaluru AI company led by co-founder and CEO Sravanth Aluru and backed by Peak XV Partners and Tiger Global.
Find more tool reviews:
