A stroke occurs when the blood supply to a region of the brain is interrupted or reduced, depriving the tissue of oxygen and nutrients. In the case of a hemorrhagic stroke, this is caused by the rupture of a cerebral vessel. Early recognition of stroke signs is essential to ensure prompt medical intervention and minimize damage to the patient, and diagnosis is typically performed through computed tomography (CT). In this study, two state-of-the-art deep learning models were integrated to accurately segment regions affected by hemorrhagic stroke in CT images. First, the action area of the stroke is delimited by the detection method YOLOV8 (You Only Look Once Version 8), which follows the Single Shot Detector (SSD) architecture to enable high-speed and high-accuracy real-time detection. Next, the Florence-2 segmentation method refines the identified region by leveraging multimodal data (text and image), thereby enhancing the clarity and precision of the segmentation. The proposed framework employs a text prompt describing the task and utilizes the polygon coordinates provided by the detection model to define the area of cerebral hemorrhage. Our results demonstrate an accuracy of 99.76%, which is comparable to other state-of-the-art methods. The robust detection capabilities of YOLOV8, combined with Florence-2’s refined segmentation, yield an efficient and adaptable approach for hemorrhagic stroke segmentation in CT images.