Anmol Agarwal

Computer Vision · GenAI · World Models

I am an undergraduate in Electrical Engineering and Computer Science at the Indian Institute of Science Education and Research (IISER) Bhopal, and a Member of Technical Staff on the Pretraining Team at Overworld, where I train foundational interactive video models and scale up data and inference pipelines for video diffusion models.

My interests span computer vision, generative AI, diffusion models, transformers, graphics programming, and simulation — with recent work centered on video world models. I am a co-author of Waypoint-1.5, a real-time video world model for consumer hardware, and of COMBAT, a conditional world model accepted at the ICCV 2025 RIWM workshop.

Previously, I was a Research Intern at the Vision and AI Lab, Indian Institute of Science (IISc), working on object relighting datasets and video diffusion models. See my CV for details, or browse my projects and publications.

news

Jul 01, 2026	Waypoint-1.5, a real-time video world model for consumer hardware, is out as a preprint. See the project page.
Oct 15, 2025	Joined Overworld as a Member of Technical Staff on the Pretraining Team, working on interactive video models.
Oct 01, 2025	Our paper COMBAT: Conditional World Models for Behavioral Agent Training was accepted at the 1st RIWM workshop at ICCV 2025. 🎉

selected publications

Preprint
Waypoint-1.5: A Real-Time Video World Model for Consumer Hardware

Rajit Rajpal, Shahbuland Matiana, Liew Wei Pyn, Anmol Agarwal, Ryan Craig, Andrew Lapp, Mithun Hunsur, Sami BuGhanem, Scottie Fox, Aaron Sanders, Carson Poole, Irene Park, David Rossi, Spencer Frazier, and Louis Castricato

2026

Preprint. * denotes equal contribution.

Abs Bib PDF Website

Waypoint-1.5 is a real-time diffusion world model for interactive video generation on consumer-grade hardware. Pre-trained on 100,000 hours of diverse, control-aligned video game data, it generates playable video conditioned on full keyboard and mouse input, with two resolution variants that run across a wide spectrum of consumer hardware.
@misc{rajpal2026waypoint, title = {Waypoint-1.5: A Real-Time Video World Model for Consumer Hardware}, author = {Rajpal, Rajit and Matiana, Shahbuland and {Liew Wei Pyn} and Agarwal, Anmol and Craig, Ryan and Lapp, Andrew and Hunsur, Mithun and BuGhanem, Sami and Fox, Scottie and Sanders, Aaron and Poole, Carson and Park, Irene and Rossi, David and Frazier, Spencer and Castricato, Louis}, year = {2026}, organization = {Overworld}, note = {Preprint. * denotes equal contribution.}, }
ICCV-W
COMBAT: Conditional World Models for Behavioral Agent Training

Anmol Agarwal, Pranay Meshram, Sumer Singh, Saurav Suman, Andrew Lapp, Shahbuland Matiana, Louis Castricato, and Spencer Frazier

In 1st Workshop on Reliable and Interactive World Models in Computer Vision (RIWM), ICCV, 2025

Non-archival

Abs Bib

COMBAT is a conditional world model that learns a reactive opponent’s tactical policy as an emergent property by training a real-time diffusion model on the 1v1 fighting game Tekken 3, demonstrating the ability to simulate an opponent that reacts to player actions.
@inproceedings{agarwal2025combat, title = {COMBAT: Conditional World Models for Behavioral Agent Training}, author = {Agarwal, Anmol and Meshram, Pranay and Singh, Sumer and Suman, Saurav and Lapp, Andrew and Matiana, Shahbuland and Castricato, Louis and Frazier, Spencer}, booktitle = {1st Workshop on Reliable and Interactive World Models in Computer Vision (RIWM), ICCV}, year = {2025}, note = {Non-archival}, }