About Me

I am a final year Ph.D. student in the College of Engineering and Computer Science (CECS) at the Australian National University, where I am supervised by Prof. Liang Zheng and Prof. Stephen Gould. I am currently a research intern at Adobe Research, working with Dr. Eli Shechtman and Dr. Richard Zhang on developing better representations for generation. I was also an intern in the Imagine-Core team at Meta GenAI under Prof. Michael Cohen, working on subject-consistent multiple-shot video generation.

Ph.D. Research Interests: Controllable image synthesis and editing, Creative content generation, Subject-consistent visual storytelling, Long-form video generation.

I also spent four beautiful summers interning at Adobe Research, working with Dr. Zhe Lin and Dr. Jianming Zhang, as well as Dr. Eli Shechtman and Dr. Richard Zhang. Prior to that, I had graduated from Masters of Machine learning and Computer Vision with top-honours from Australian National University. Even earlier, I was worked as machine learning research engineer at Yahoo Japan. I received my Bachelors in Electrical Engineering with specialization in intelligent and cognitive systems from IIT Delhi.

News

[May, 2026] Introducing RAEv2 🚀 - Improved baselines with Representation Autoencoders: over 10x faster convergence with better reconstruction and generation!
[April, 2026] iREPA accepted to ICLR 2026 🎉🎉🎉
[Dec, 2025] Introducing iREPA 🎨 - What matters for representation alignment: global information or spatial structure?
[Oct, 2025] Introducing REPA-E T2I 🚀 - Family of end-to-end trained latent diffusion models for text-to-image generation with >5x faster training!
[July, 2025] Introducing DeepSWE-Preview 🚀 - A reasoning-enabled coding agent achieving 59.0% on SWE-Bench-Verified with test-time scaling, setting new SOTA for open-weight coding agents!
[July, 2025] R2E-Gym accepted to COLM 2025 (Top 5 Papers) and NeurIPS Workshop 2025 🎉
[April, 2025] Released R2E-Gym 🚀 - Achieved 51% success rate on SWE-Bench, setting new SOTA for open-weight SWE agents!
[April, 2025] Released REPA-E 🎨 - Enabling stable end-to-end training of latent diffusion transformers with SOTA generation performance and 45x accelerated training!
[April, 2025] REPA-E accepted to ICCV 2025 🏝️ See you in Hawaii!! 🌺
[March, 2025] 1 paper accepted to CVPR 2025 🎉
[Jan, 2025] Three papers (1 Oral and 2 Poster) accepted at ICLR 2025 🎉🎉🎉
[May, 2024] Started Internship at Meta GenAI working on subject-consistent video generation 🤗

Selected Publications

	Improved Baselines with Representation Autoencoders Jaskirat Singh, Boyang Zheng, Zongze Wu, Richard Zhang, Eli Shechtman, Saining Xie ¹Adobe Research, ²Australian National University, ³New York University 🚀 Over 10x faster convergence with better reconstruction, generation, and improved T2I and world model performance! [Paper] [Code] [Project Website] [🤗 Models]
	What Matters for Representation Alignment: Global Information or Spatial Structure? Jaskirat Singh, Xingjian Leng, Zongze Wu, Liang Zheng, Richard Zhang, Eli Shechtman, Saining Xie ¹Adobe Research, ²Australian National University, ³New York University ICLR 2026 [Paper] [Code] [Project Website] [🤗 Models]
	REPA-E T2I: Family of End-to-End Trained Latent Diffusion Models for Text-to-Image Generation REPA-E Team: Xingjian Leng, Jaskirat Singh,** Yunzhong Hou, Zhenchang Xing, Saining Xie, Liang Zheng Canva Team: Ryan Murdock, Ethan Smith, Rebecca Li *Project Lead (Equal Contribution) ¹Australian National University, ²Data61-CSIRO, ³New York University, ⁴Canva 🚀 >5x faster T2I training without additional representation alignment loss! [Paper] [Code] [Project Website] [Models]
	DeepSWE-Preview: Training a Fully Open-sourced State-of-the-Art Coding Agent by Scaling RL Agentica: Michael Luo, Jaskirat Singh,** Naman Jain, Sijun Tan, Colin Cai*, Tarun Venkat, Manan Roongta, Li Erran Li, Raluca Ada Popa, Koushik Sen, Ion Stoica Together AI: Ameen Patel, Qingyang Wu, Alpay Ariyak, Shang Zhu, Ben Athiwaratkun, Ce Zhang ¹UC Berkeley, ²Together AI 🏆 SOTA reasoning-enabled coding agent using only RL!! Achieves open-weight SOTA 59.0% on SWE-Bench-Verified (42.2% Pass@1, 71.0% Pass@16)! [Project Page] [Code] [🤗 Model] [🤗 Dataset] [Eval Logs]
	R2E-Gym: Procedural Environment Generation and Hybrid Verifiers for Scaling Open-Weights SWE Agents Naman Jain, Jaskirat Singh,** Manish Shetty, Liang Zheng, Koushik Sen, Ion Stoica *Equal Contribution ¹UC Berkeley, ²Australian National University COLM 2025 (Top 5 Papers), NeurIPS Workshop 2025 [Paper] [Code 🤗] [Project Website] [Data & Models]
	REPA-E: Unlocking VAE for End-to-End Tuning of Latent Diffusion Transformers Xingjian Leng, Jaskirat Singh,** Yunzhong Hou, Zhenchang Xing, Saining Xie, Liang Zheng *Project Lead (Equal Contribution) ¹Australian National University, ²Data61-CSIRO, ³New York University ICCV 2025 [Paper] [Code] [Project Website] [🤗 Models]
	Negative Token Merging: Image-based Adversarial Feature Guidance Jaskirat Singh, Lindsey Li, Weijia Shi, Ranjay Krishna, Yejin Choi, Pang Wei Koh, Michael F. Cohen, Stephen Gould, Liang Zheng, Luke Zettlemoyer. ¹University of Washington, ²Allen Institute for AI, ³Australian National University 🏆 Image based classifier-free guidance [instead of text] to improve output diversity, quality, control in just few lines of code! [Paper] [Code] [Project Website] [Huggingface Demo 🤗]
	OpenDevin: An Open Platform for AI Software Developers as Generalist Agents Best LLM Agent [🥇 SWE-Bench-Lite] [🥇 SWE-Bench-Verified] ICLR 2025: Oral [Paper] [Code] [Project Website]
	Storybooth: Training-free Multi-Subject Consistency for Improved Visual Storytelling Jaskirat Singh, Junshen Kevin Chen, Jonas Kohler, Michael Cohen ¹Meta GenAI, ²Australian National University ICLR 2025 [Paper]
	Vec2Face: Scaling Face Dataset Generation with Loosely Constrained Vectors Haiyu Wu, Jaskirat Singh, Sicong Tian, Liang Zheng, Kevin Bowyer. ICLR 2025 [Paper] [Code] [Project Website]
	SmartMask: Context Aware High-Fidelity Mask Generation for Fine-grained Object Insertion and Layout Control Jaskirat Singh, Jianming Zhang, Qing Liu, Cameron Smith, Zhe Lin, Liang Zheng ¹Adobe Research, ²Australian National University CVPR 2024, US Patent [Paper] [Project Website]
	Divide, Evaluate, and Refine: Evaluating and Improving Text-to-Image Alignment with Iterative VQA Feedback Jaskirat Singh and Liang Zheng. NeurIPS 2023 [Paper] [Code] [Project Website]
	IMPUS: Image Morphing with Perceptually-Uniform Sampling Using Diffusion Models ICLR 2024 [Paper] [Code]
	High-Fidelity Guided Image Synthesis with Latent Diffusion Models Jaskirat Singh, Stephen Gould, and Liang Zheng. CVPR 2023 [Paper] [Code] [Project Website]
	Paint2Pix: Interactive Painting based Progressive Image Synthesis and Editing Jaskirat Singh, Liang Zheng, Cameron Smith, Jose Echevarria ¹Adobe Research, ²Australian National University ECCV 2022 [Paper] [Code] [Project Website]
	Intelli-Paint: Towards Developing More Human-Intelligible Painting Agents Jaskirat Singh, Cameron Smith, Jose Echevarria, Liang Zheng ¹Adobe Research, ²Australian National University ECCV 2022, US Patent [Paper] [Project Website]
	Combining Semantic Guidance and Deep Reinforcement Learning for Generating Human Level Paintings Jaskirat Singh and Liang Zheng. CVPR 2021 [Paper] [Code]
	Enhanced Scene Specificity with Sparse Dynamic Value Estimation Jaskirat Singh and Liang Zheng arXiv [Paper]
	Dynamic Value Estimation for Single-Task Multi-Scene Reinforcement Learning Jaskirat Singh and Liang Zheng arXiv [Paper]