NoiseCollage is a novel layout-aware text-to-image diffusion model that addresses the challenges of text and layout condition mismatches and image quality degradation in existing models. The model independently estimates noises for individual objects and then crops and merges them into a single noise, enabling accurate placement of objects in their correct locations. This approach outperforms several state-of-the-art models in both qualitative and quantitative evaluations. NoiseCollage can be integrated with ControlNet to use additional conditions such as edges, sketches, and pose skeletons, enhancing layout accuracy. The model is training-free, allowing it to leverage pre-trained diffusion models for image generation. Experiments show that NoiseCollage generates high-quality images with accurate layout and text conditions. The model's crop-and-merge operation of noises is a reasonable strategy for controlling image generation. However, it sometimes fails to generate small objects and may create images with fake relationships between people. Future work will focus on improving layout control and exploring the properties of noise representation.NoiseCollage is a novel layout-aware text-to-image diffusion model that addresses the challenges of text and layout condition mismatches and image quality degradation in existing models. The model independently estimates noises for individual objects and then crops and merges them into a single noise, enabling accurate placement of objects in their correct locations. This approach outperforms several state-of-the-art models in both qualitative and quantitative evaluations. NoiseCollage can be integrated with ControlNet to use additional conditions such as edges, sketches, and pose skeletons, enhancing layout accuracy. The model is training-free, allowing it to leverage pre-trained diffusion models for image generation. Experiments show that NoiseCollage generates high-quality images with accurate layout and text conditions. The model's crop-and-merge operation of noises is a reasonable strategy for controlling image generation. However, it sometimes fails to generate small objects and may create images with fake relationships between people. Future work will focus on improving layout control and exploring the properties of noise representation.