Alibaba Wan 2.7 Image Edit (Official Edition)

Text-to-image is great until you need to change one thing in an image you already like. That's the case for instruction-based editing: load the picture, type "turn this into a rainy evening, keep the composition," and the model does it instead of you reaching for an inpaint mask and praying. This node is Wan 2.7's take on that, running on Alibaba's cloud through the BizyAirPlus pack - no local model, no GPU, just a prompt and an API key.

It's the image-editing sibling of the Wan 2.7 T2I node in the same pack, and it shares the family DNA: you edit the graph locally, flip the BizyAirPlus button ON, and the heavy lifting happens server-side.

How it works

You feed it images (the required images input is an autogrow slot, so it happily takes several) and size, and describe the edit in prompt. The tooltip sets up the referencing convention: talk about your inputs as "Figure 1", "Figure 2", and so on, in the order they're attached. "Figure 1: change the sky to a sunset." That's the whole interface.

The outputs are image, request_id, and urls. Editing is capped at 1K and 2K - there's no 4K tier here, which the size tooltip says plainly ("image editing scenarios are supported at 1K and 2K, not 4K"). When you pass images, the output aspect ratio follows the last image.

The inputs that matter

• prompt - the edit command. ≤5000 characters, Chinese or English, and reference inputs by figure number.
• images - one or more source pictures. This is where the edit happens.
• size - 1K or 2K; output ratio follows the last input image.
• bbox_list - the power-user knob. A JSON array of box selections, same length as the number of images, each box [x1, y1, x2, y2] in absolute pixels of the original, up to two boxes per image. This is how you tell the model where in the frame to work - essential for surgical edits.
• color_palette - a JSON color theme (3–10 hexes with ratios summing to 100%) when you want to restyle toward a palette.
• custom_width / custom_height - only when size is Custom, and the tooltip warns of a 2048×2048 total-pixel ceiling for editing.

Plus watermark, skip_error, and api_config (per-request endpoint/key override).

Installation

The one pack install, done once. ComfyUI Manager, search BizyAirPlus, or:

cd ComfyUI/custom_nodes
git clone https://github.com/siliconflow/BizyAirPlus.git
cd BizyAirPlus
python -m pip install -r requirements.txt

Restart ComfyUI, then add a bizyair.ai API key (ON-button prompt, Settings > BizyAirPlus > API Key, or BIZYAIR_API_KEY).

Common issues

• Edit affects the whole image when you wanted a region - use bbox_list to scope the edit to a box. The prompt alone tends to work globally.
• bbox rejected - the list must match the number of input images, boxes are absolute pixel coordinates, and you get two per image max.
• "4K" isn't available - by design for editing. 2K is the ceiling on this node.
• Cloud failures - key and network first, then the ComfyUI console for the detailed error.