WD1.4, florence2, MiniCPMZ automatically tag.

Training data labeling workflow can use the w1.4 model for keyword tagging, or Microsoft's latest on-device model Florence2 for natural language tagging, as well as the powerful domestic on-device model MiniCPM for natural language tagging. The latter two are suitable for generating Flux training data, while WD1.4 is suitable for all training.

If it is for character optional cut-out, the cut-out models can use models like RMBG1.4 or U2NET, which can be downloaded independently.