Nanobanana 提示词工程工作流

① Create a "{argument name="character 1" default="Polar Bear"}" and a "{argument name="character 2" default="Panda"}" wearing ice skates, and the background location. Normal chat-based input is also OK. Whether you generate it all at once or separately, the process is the same. ② Perform annotation on the image. This is the most troublesome part, so I made it a tool. Based on the file name, I create a prompt that organizes "which part of which image to reference." (Pose / BG / Face / Clothing, etc.) ③ Add supplementary information such as composition and color scheme for the final generation.