Trendy diffusion-based picture generative fashions have made vital progress and turn into promising to complement coaching knowledge for the article detection process. Nevertheless, the era high quality and the controllability for complicated scenes containing multi-class objects and dense objects with occlusions stay restricted. This paper presents ODGEN, a novel technique to generate high-quality photographs conditioned on bounding bins, thereby facilitating knowledge synthesis for object detection. Given a domain-specific object detection dataset, we first fine-tune a pre-trained diffusion mannequin on each cropped foreground objects and whole photographs to suit goal distributions. Then we suggest to regulate the diffusion mannequin utilizing synthesized visible prompts with spatial constraints and object-wise textual descriptions. ODGEN displays robustness in dealing with complicated scenes and particular domains. Additional, we design a dataset synthesis pipeline to judge ODGEN on 7 domain-specific benchmarks to show its effectiveness. Including coaching knowledge generated by ODGEN improves as much as 25.3% mAP@.50:.95 with object detectors like YOLOv5 and YOLOv7, outperforming prior controllable generative strategies. As well as, we design an analysis protocol based mostly on COCO-2014 to validate ODGEN basically domains and observe a bonus as much as 5.6% in mAP@.50:.95 in opposition to present strategies.