CountLoop: Training-Free High-Instance Image Generation via Iterative Agent Guidance
This addresses the challenge of precise instance control in image generation for applications requiring high-density scenes, though it is incremental as it builds on existing diffusion models.
The paper tackled the problem of diffusion models being unreliable for generating images with a precise number of object instances in complex settings, and presented CountLoop, a training-free framework that uses iterative agent feedback to achieve up to 98% counting accuracy while maintaining visual quality.
Diffusion models have shown remarkable progress in photorealistic image synthesis, yet they remain unreliable for generating scenes with a precise number of object instances, particularly in complex and high-density settings. We present CountLoop, a training-free framework that provides diffusion models with accurate instance control through iterative structured feedback. The approach alternates between image generation and multimodal agent evaluation, where a language-guided planner and critic assess object counts, spatial arrangements, and attribute consistency. This feedback is then used to refine layouts and guide subsequent generations. To further improve separation between objects, especially in occluded scenes, we introduce instance-driven attention masking and compositional generation techniques. Experiments on COCO Count, T2I CompBench, and two new high-instance benchmarks show that CountLoop achieves counting accuracy of up to 98% while maintaining spatial fidelity and visual quality, outperforming layout-based and gradient-guided baselines with a score of 0.97.