This workflow uses qwen3 vl to create a 10 prompt list each time you run it and qwen edit 2509 to generate the images. If you want the wan 2.2 refinement, leave the group on. This workflow is perfect for generating the perfect dataset to train a lora.