This is a image to image workflow.

This workflows using SAM3 or YOLO to do auto detection, and create mask as filter, change the image to realist while keeping part of it to orginal. We can use it to keep like character or face in anime style.

Please note that YOLO may NOT having a wide scope for many words, but it is faster and using lower VRAM. You may check the prompt in YOLO offical documents.

You NEED to download the base model (flux-2-kelin-9B-base) and the lora (flux-2-kelin-anime2realist) to use it. The links are in the Suggested Resources below.