Alibaba Wan 2.7 Video Edit (Official Edition)

Most video generation is a one-shot: write a prompt, pray, re-roll. Video editing is the grown-up version where you already have footage and want to change what's happening in it - restyle the whole clip, swap the setting, make a character do something else - without regenerating from scratch. That's what this node does, running Wan 2.7's "edit by instruction" capability on Alibaba's cloud. It's part of the BizyAirPlus pack, so it needs no local model, no GPU, just an API key and a source video.

It's the closest thing in the pack to a "prompt engineer's tool": the whole job is describing the edit precisely. The node's prompt tooltip draws the line cleanly - pure instructions describe style changes and similar global edits, while reference diagrams describe substitution or modification operations (putting a specific object in, taking one out). Keep that distinction in your head and the node behaves.

How it works

Attach a source video, and the node sends it plus your instructions to the Wan 2.7 cloud endpoint. The required inputs are resolution and audio_setting, which defaults to auto: the model decides from the prompt whether to regenerate audio or keep the original sound. Set it to origin to force the original audio through. That's a detail most similar tools don't even expose, and it's the difference between a reskin and a full re-sound.

Every run returns video (into SaveVideo), request_id, and urls.

The inputs that matter

• prompt - up to 5000 characters of edit instructions. Be concrete: "make this a rainy night scene, keep the same camera movement" beats "change the mood."
• duration - 0 keeps the input video's length (and isn't passed to the API at all); 2–10 trims the output to that many seconds from the start. Default 5, and it directly affects cost.
• ratio - defaults to default, which simply follows the input video and isn't sent to the API. Choose a specific value only if you want to force an output ratio.
• audio_setting - auto (smart choice from the prompt) vs origin (keep original sound).
• negative_prompt - ≤500 chars, describes what shouldn't appear in the output frames.

Plus seed (-1 = server-random), watermark, prompt_extend (LLM smart-overwrite for short prompts), skip_error to keep the graph alive on failure, and api_config to override the endpoint/key per request.

Installation

The standard BizyAirPlus install, done once for the whole pack. 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 via the ON-button prompt, Settings > BizyAirPlus > API Key, or the BIZYAIR_API_KEY environment variable. No local model files.

Common issues

• Edit does nothing / too subtle - your prompt is reading as a pure description, not an instruction. Say what changes, explicitly. If you want an object swapped in, attach it as a reference diagram.
• Audio came out wrong - that's audio_setting on auto making a call from your prompt. Force origin if you must keep the source sound.
• Output length unexpected - duration of 0 means "keep input length"; anything else trims from 0 seconds. Confirm which you set.
• Cloud failures - check the key and network, then the ComfyUI console for the detailed error.