<p>Wildfires are a common occurrence worldwide and cause loss of life and economic damage every year. Machine learning models have become an important mechanism for fire prevention and detection. Among these technologies, wildfire segmentation can help detect precise wildfire areas and improve fire-fighting capabilities. Numerous techniques have been introduced for segmenting wildland fire zones, such as convolutional neural networks. However, these models are sensitive to interruptions caused by noise. Recently, diffusion models showed interesting results for segmenting medical images. In this paper, we introduced a novel diffusion model based wildfire segmentation method, namely DM-Fire. This method employs a median filter and the probabilistic diffusion model MedSegDiff for generating wildfire segmentation masks. The proposed DM-Fire achieved promising results with an F1-score of 91.06% better than existing models. This demonstrates the potential of diffusion models in segmenting forest fire images and addressing fire challenging, including the detection of small fire areas and background complexity as well as improving fire management strategies.</p>