In a significant leap for India’s burgeoning artificial intelligence sector, Avataar AI has introduced Varya, an innovative video generation model tailored to meet the nation’s diverse cultural contexts at an unprecedented cost. Priced at a mere ₹0.48 ($0.005) per second, Varya promises to democratize video content creation, making it accessible to students, small businesses, and larger enterprises alike.
This development comes in the wake of India's AI ambitions, which have historically lagged behind those of the U.S., Europe, and China. Recognizing the need for rapid advancement, the Indian government launched the India AI Mission, a $1.2 billion initiative designed to bolster local startup capabilities by providing subsidized GPU computation in exchange for open-sourcing their models.
Avataar AI is one of the twelve startups selected for this initiative, leveraging its support to refine Varya, which is built upon the foundations of Alibaba’s Wan 2.2 model. Through a process known as model distillation, Avataar has successfully engineered a version that operates in just four steps—compared to Wan 2.2's cumbersome fifty—yielding video generation ten times faster and at a fraction of the cost. To illustrate the efficiency of Varya, using an NVIDIA H200 GPU, a 5-second 720p video can be created in approximately 45 seconds, as opposed to the substantial 1,230 seconds required by its predecessors.
The affordability of Varya is set to disrupt the current landscape dominated by video AI models like Veo, Kling, Luma, and Runway, which command rates of $0.10 and upwards per second. Rajan Anandan, Managing Director at Peak XV and a key investor in Avataar, emphasized, “India is a video-first market. Current AI video models are too expensive for population-scale use here. Reducing costs dramatically is essential for widespread adoption of video AI across various sectors.”
Apart from its cost efficiency, Varya boasts an integrated understanding of local customs and cultures, crucial for generating authentic and contextually relevant content. Avataar AI has curated specific datasets to train Varya, ensuring it recognizes essential cultural elements, such as festivals, attire, food, and architecture. This attention to cultural detail separates Varya from existing models, which often produce generic outputs devoid of local flavor.
In a bid to foster further innovation, Avataar plans to release Varya as an open-weight model via the India AI Kosh portal—an initiative to assemble publicly available AI tools and data. This transparency will allow developers to either self-host or adapt the model for various applications. The company also aims to collaborate with established video tools like Adobe Firefly and Higgsfield, enhancing Varya's applicability across different platforms.
This bold launch encapsulates a broader strategic vision within India’s AI landscape. Industry analysts suggest that India can carve a niche for itself by focusing on application development as opposed to competing with global leaders on foundational models. Given challenges such as limited computational resources and access to quality data, the India AI Mission represents a crucial step toward bridging this gap, targeting $200 billion in AI investments by 2028 while aiming to double GPU capacity in the near term.
As the digital landscape continues to evolve, Varya stands poised to redefine video content creation across India, heralding a new era of creativity, accessibility, and cultural richness.
Source: TechCrunch
