Trigger scenarios
Examples of user wording:
- Style transfer: "Change this image to ink wash painting style", "turn it into anime style"
- Background replacement: "Change the background to white", "change the indoor background to the seaside"
- Content modification: "Change the cat to a dog", "add some falling cherry blossoms to the scene"
- Outpainting generation: "Extend the scene to the right"
Out of scope
- Images generated from scratch (no original image) → not within the scope of this skill, handled by image-gen-skill
- Text generation requests → not within the scope of this skill, handled by text-gen-skill
- Professional retouching that requires pixel-level precision → not within the scope of this skill, AI cannot guarantee pixel-level precision
Interface chain
editImage: based on the original image and editing requirements, achieve the image editing effect through AI redrawing
Usage order
- The user needs to provide an original image (image URL) and an editing description
- Generate a new prompt based on the editing description, and call the image generation model to regenerate
- Return a comparison between the original image and the edited image
- The edited image is automatically persisted to cloud storage
Design constraints
- This is editing implemented through AI redrawing, not pixel-level precise editing, and it cannot guarantee complete consistency with the original image
- It must be called through a cloud function; the mini program side does not directly process images
- The original image needs to provide an accessible network URL