This is i2i's workflow using Qwen.

The prompt supports both automatic and manual input from images.