Skin diseases are one of the most prevalent health concerns worldwide, affecting individuals of all ages and backgrounds. Even though most of the diseases are not causing mortality, it potentially decreases quality of life. Hence appropriate diseases management should be delivered. In some rural countries such as Indonesia, the distribution of dermatologist is imbalance which mostly concentrated at big cities. Hence, dermatological healthcare is mostly conducted by general practitioner for most rural areas. Due to the limitation of dermatological competency by the general practitioner, some misdiagnosis is inevitable. Artificial Intelligence support through machine learning modeling has been developed to provide prediction model-based technology that can support doctors on confirming diagnosis. However, one of the challenges on building the predicting model is the insufficient annotated skin diseases image dataset. This paper proposes and analyses the potential of the few shots learning framework using prototypical network architecture to construct the skin disease prediction model dealing with the few numbers of an-notated skin diseases image dataset. The simulation shows that the proposed approach is promising to be further developed on prediction modeling for skin disease prediction based on spot diagnosis based on its consistency on the given performance parameter and the ability to reduce the required annotated data to less than one (1) percent.