About

I am a PhD student at Inria Paris in the Astra-Vision group, advised by Raoul de Charette and Fabio Pizzati. My research focuses on physics-aware world models: making video generation models encode physical properties and simulate real-world dynamics.

Before this, I was a Research Assistant at the Vision and AI Lab (VAL), Indian Institute of Science (IISc), Bangalore, advised by R. Venkatesh Babu, and prior to that a Senior Engineer at Samsung R&D Institute Bangalore. I completed my MS (By Research) in Computer Science and Engineering at IIT Kanpur, advised by Gaurav Sharma, where my thesis focused on audio-guided facial expression editing using StyleGAN.

Outside research, I enjoy playing guitar 🎸 and singing 🎵.

News

Publications · click cards to expand

Principia overview
Principia: Relational Physics Tests for Video Models
NeurIPS 2026 E&D Track
Varun Varma Thozhiyoor*, Shivam Tripathi*, R. Venkatesh Babu, Anand Bhattad *Equal contribution
TL;DR Modern video generators fail surprisingly simple Newtonian physics tests. Principia uses calibration-independent relational constraints and finds no generator exceeds 0.5 physical consistency despite all scoring >0.7 on VBench.
Paper Project Page
Click to expand
Objects in Generated Videos Are Slower Than They Appear
CVPR Findings 2026
Varun Varma Thozhiyoor, Shivam Tripathi, R. Venkatesh Babu, Anand Bhattad
TL;DR Video diffusion models exhibit "sub-Earth" gravity — objects fall up to 4× slower than physics dictates. A LoRA fine-tuned on just 100 clips raises effective gravity from 1.81 → 6.43 m/s² (65% of Earth's g).
Paper Project Page
Click to expand