Granulomatous rosacea (GR) and lupus miliaris disseminatus faciei (LMDF) exhibit overlapping clinical features, making their differentiation challenging. While histopathological examination remains the gold standard, it is invasive and time-consuming, highlighting the need for non-invasive diagnostic approaches. This study evaluates artificial intelligence (AI)-based models for differentiating between GR and LMDF and assess their impact on clinician performance. This retrospective pilot study included 96 patients (62 GR, 34 LMDF) with histopathologically confirmed diagnoses. Neural network models, including convolutional neural networks and vision transformers (ViT), were applied to cropped lesion images while a transformer-based multiple instance learning (TransMIL) approach was used for whole-image analysis. Diagnostic accuracy was also compared between clinicians with and without AI assistance. ViT_base_patch16_224 achieved the highest accuracy (93.0%) and reliability (κ = 0.81) on cropped images, while the TransMIL reached 70% accuracy on whole images. AI augmentation significantly improved clinicians' diagnostic accuracy from 64.7% to 70.3% (p = 0.0136), with the greatest improvement observed among general practitioners. Additionally, mean diagnostic time decreased from 10.7 to 6.4 minutes. These findings highlight the potential of AI models, particularly ViT, in facilitating the differential diagnosis of GR and LMDF. AI-augmented diagnosis improved accuracy and efficiency across all clinician expertise levels, supporting its integration as a complementary tool in dermatological practice.