Flex-angle detection of objects or human bodies has potential applications in robotic arm control, medical rehabilitation, and deformation detection. However, current solutions such as flex sensors and computer vision methods have limitations, such as the limited system lifetime of battery-powered flex sensors and the failure of computer vision methods in Non-Line-of-Sight (NLoS) scenarios. To overcome these limitations, we propose an RFID-based flex-sensor system, called RFlexor, which enables batteryless flex-angle detection in NLoS scenarios. RFlexor utilizes the change in tag phase and Received Signal Strength Indicator (RSSI) caused by the flexing of the tag to detect flex-angles. To extract the complex relationship between tag phase/RSSI and flex-angle, we train a multi-input AI model. We address the significant technical challenges by reformulating the phase and RSSI models, using phase difference and RSSI ratio as inputs, and applying multi-head attention to fuse phase and RSSI data. We implement the RFlexor system using Commercial-Off-The-Shelf (COTS) RFID devices and conduct extensive experiments. The results show that RFlexor achieves fine-grained flex-angle detection with a detection error of fewer than 10 degrees and a probability higher than 90% in most conditions. The average detection error is always less than 10 degrees across all experiments. Overall, RFlexor provides a promising solution for flex-angle detection in various scenarios.