Alibaba Wan 2.7 Image Edit Pro (Official Edition)

If the base Wan 2.7 image-edit node is the workhorse, this is the higher-tier edition of the same tool. Same idea - load images, type an instruction, get an edited result - running on the Pro model tier inside the BizyAirPlus pack's cloud. Where the base node is capped at 2K output, so is this one; the difference between the editions lives in the model quality behind the API, not in a wall of extra knobs.

The schema is effectively a mirror of the base I2I node: required images (autogrow, so several can feed in) and size (1K/2K - the tooltip is explicit that editing doesn't do 4K), with prompt, color_palette, bbox_list, custom_width, custom_height, watermark, skip_error, and api_config as options. Outputs are image, request_id, and urls.

How it works

Nothing changes about the mechanics: reference your inputs as "Figure 1" and "Figure 2" in the prompt (≤5000 chars, bilingual), and use bbox_list - a JSON array of [x1, y1, x2, y2] pixel boxes, one per input image, two boxes per image max - to scope the edit to a region. Output aspect ratio follows the last input image. The Pro difference is the model tier under the hood: same surgery, better results, usually at slightly higher cost per request. That's the honest pitch.

The inputs that matter

• prompt - the edit instruction, with figure-number referencing.
• images - your source material, one or more.
• size - 1K or 2K. Remember: no 4K for editing, Pro or not.
• bbox_list - the surgical box selection. This is the input that separates "change the sky" from "change just that sky."

Installation

One-time, for the whole pack:

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

Or ComfyUI Manager, search BizyAirPlus. Restart, add a bizyair.ai API key, and flip the BizyAirPlus button ON in the action bar.

Common issues

• Paying more for the same cap - if 2K is fine and your edits aren't demanding, the base node is the cheaper call. Reach for Pro when the base edition's edits feel mushy.
• Edits going global - scope with bbox_list; the model reads whole-frame changes when no box is given.
• bbox rejected - length must match the image count, coordinates are absolute pixels, two boxes per image.
• Cloud failures - key and network first, then the ComfyUI console.