CountLoop: Training-Free High-Instance Image Generation via Iterative Agent Guidance

TL;DR

COUNTLOOP is a training-free method that uses iterative feedback from vision-language models to generate images with precise object counts and high spatial quality, especially in dense scenes.

Contribution

It introduces a novel training-free framework combining scene layout planning and feedback-driven refinement for high-instance image generation.

Findings

Reduces counting error by up to 57% on benchmarks.

Achieves highest or comparable spatial quality scores.

Maintains photorealism in densely occluded scenes.

Abstract

Diffusion models excel at photorealistic synthesis but struggle with precise object counts, especially in high-density settings. We introduce COUNTLOOP, a training-free framework that achieves precise instance control through iterative, structured feedback. Our method alternates between synthesis and evaluation: a VLM-based planner generates structured scene layouts, while a VLM-based critic provides explicit feedback on object counts, spatial arrangements, and visual quality to refine the layout iteratively. Instance-driven attention masking and cumulative attention composition further prevent semantic leakage, ensuring clear object separation even in densely occluded scenes. Evaluations on COCO-Count, T2I-CompBench, and two newly introduced high instance benchmarks show that COUNTLOOP reduces counting error by up to 57% and achieves the highest or comparable spatial quality scores across all benchmarks, while maintaining photorealism.