It was accepted to the ACM Transactions on Graphics (ACM-TOG), 2025, the flagship journal of the ACM SIGGRAPH community. Acceptance to ACM-TG represents a mark of exceptional quality and places the work among the most impactful contributions in computer graphics and vision research. Over 8,000 attendees from 60 countries and regions came together to learn, network, and do business at one of the region's most established Computer Graphics events.

Research Paper: MALeR: Improving Compositional Fidelity in Layout-Guided Generation Text-to-image generation models can create truly remarkable images. However, these models struggle in generating images involving multiple subjects. Providing layout-guidance is useful since it allows spatial control over generated images but often results in multiple failure cases. “Our paper, MALeR, addresses these failure cases and allows users to generate high-quality compositional scenes involving multiple subjects with multiple attributes such as colors and materials.” Said Shivank.

MALeR addresses key limitations in existing methods by enhancing the alignment between generated content and prescribed layouts, enabling more accurate placement, relationships, and visual coherence of objects in complex scenes. This work contributes meaningfully to advancing controllable and reliable generative models — a growing area of interest across graphics, vision, and AI.

 

Presentation at SIGGRAPH Asia

Presentation at SIGGRAPH

 

Sharing Author’s Experience:

We presented our paper at SIGGRAPH ASIA, held in Hong Kong in December 2025. Our group, Katha AI, led by Prof. Makarand Tapaswi, focuses primarily on video understanding and this was our group’s first foray into a major graphics venue. We were extremely grateful to have received a journal track recommendation from the conference committee, especially since this year’s ACM TOG journal track recommendation rate was just 4% of the accepted conference papers.

SIGGRAPH ASIA was a total roller-coaster ride for us, mostly fun and exhilarating, with a couple of overwhelming moments. The conference started with a 20-second fast-forward presentation of our work in front of a packed hall. Following that, giving a detailed 10-minute talk before stalwarts in the field of generative modelling was intimidating, yet an extremely memorable moment. The conference was truly a visual extravaganza, from interactive VR/AR experiences to stunning real-time tech demos. The conference also showcased multiple animated short films presented by directors from across the globe. Technical talks were a major highlight, where we got to learn about emerging research trends in generative modelling, graphics and robotics. It was truly a remarkable learning experience on every front. Finally, none of this would have been possible had it not been for Makarand sir’s guidance throughout, and we consider ourselves very lucky to be under his guidance. His support played a crucial role in shaping both the work and the confidence to present it on a global stage. The presentation offered valuable exposure to cutting-edge feedback and discussions within the global graphics research community.

 

A Notable Achievement

The acceptance of this work to ACM Transactions on Graphics underscores the strength and originality of the research, as well as Shivank and Dhruv’s contribution to the field of generative modeling and computer graphics. Presenting at SIGGRAPH Asia further highlights the global relevance and impact of the work.

This achievement reflects the growing research excellence emerging from young researchers and reinforces the institute’s presence in top-tier international venues.