Think, Plan, Paint: Layout-Aware Reasoning for Controllable Image Generation in Unified Models

2026-07-17

AI Summary

Existing unified multimodal large language models struggle to achieve controllable image generation under complex spatial instructions and logical constraints. This work proposes ATLAS, a novel framework that introduces, for the first time in a unified model, a human-like “think–plan–draw” three-stage paradigm, using layout as a shared representation to jointly perform spatial reasoning, object arrangement planning, and image rendering. The framework supports layout alignment, instruction editing, and multimodal grounding, and introduces the ATLAS-Reasoning evaluation benchmark. Experimental results demonstrate that ATLAS significantly outperforms current state-of-the-art methods in image generation, achieving an average improvement of 65.31% over the best existing layout-based unified models, and a 23.06% gain on spatial reasoning tasks.

Abstract

Unified Multimodal Large Language Models (MLLMs) offer a promising paradigm for unifying visual understanding and generation, yet they still struggle to follow complex spatial instructions and logical constraints in controllable image generation. To address this gap, we present ATLAS, a unified framework that equips MLLMs with a human-like "Think, Plan, and Paint" paradigm. We adopt layout as the shared representation that connects the three stages, enabling the model to reason about spatial requirements, plan explicit object arrangements, and render the final image. We further improve plan-to-image fidelity with reinforcement-learning-based layout alignment. We instantiate ATLAS at 7B and 80B scales, achieving state-of-the-art performance among MLLMs on image generation benchmarks and an average 65.31% improvement over existing layout-based unified MLLMs. On spatially related tasks, ATLAS obtains an average 23.06% gain over the base models. Through the same layout interface, ATLAS also supports instruction-guided editing and multimodal grounding. We further introduce ATLAS-Reasoning, a benchmark for evaluating generation under complex spatial instructions.

Problem

• controllable image generation
• spatial instructions
• logical constraints
• multimodal large language models
• layout-aware reasoning

Innovation

• layout-aware reasoning
• controllable image generation
• unified multimodal LLMs
• reinforcement learning alignment
• spatial instruction following