Wan 2.7 海外图像生成/编辑（3 合 1）

Here's the context you need before touching this node: Wan 2.2 was the last open-weights Wan. Everything after it - 2.5, 2.6, 2.7, and 3.0 - is API-only from Alibaba, so the community that used to download Wan now has to call it. Wan_2_7_Global_Image is that call, for the overseas ("global") image line specifically, and it bundles three models under one dropdown: wan-2.7-global-t2i for text-to-image, plus wan-2.7-global-i2i and wan-2.7-global-i2i-pro for image editing with up to nine reference images.

The "overseas" bit is worth understanding because it's a real thing in this pack: T8mars routes domestic and global versions of the same models through different upstreams (you see the same split on the Wan 3.0 video node). The global line is the one to reach for if the domestic endpoint is slow, blocked, or has harsher review for what you're doing.

How it works

Like every node in this pack, it's an HTTP client with nice packaging. Your prompt goes to the reseller's /v1/image/generations endpoint, local reference images get uploaded automatically, and the node polls the task, downloads the result, and returns a decoded ComfyUI IMAGE. Nothing runs on your GPU - the only local dependency is requests, and the only cost is the per-image API call.

The model choice decides which inputs matter. For t2i, you get width/height (512–4096, step 8) and a thinking_mode toggle that enables the upstream reasoning mode. For the two i2i models, the width/height and thinking controls are ignored and the image1–image9 reference slots take over instead - connect at least one, describe what you want changed, and the model edits. The prompt cap differs too: 5000 characters for T2I, 2048 for I2I, so don't paste your thousand-word lore dump into an edit job.

Inputs and outputs that matter

• model - wan-2.7-global-t2i, wan-2.7-global-i2i, or wan-2.7-global-i2i-pro.
• prompt - the creative input; T2I up to 5000 chars, I2I up to 2048.
• width / height - T2I only, 512–4096 in steps of 8.
• thinking_mode - T2I only; turns on upstream reasoning.
• image1–image9 - I2I references, ignored in T2I.

Outputs: image, image_url, task_id, response - standard for the pack's image nodes.

Install

This is one node inside T8mars/ComfyUI_Seedance, so installing the pack gets you this and every other node in the family:

# ComfyUI Manager: search "ComfyUI Seedance", or:
comfy node install seedance

Or git clone https://github.com/T8mars/ComfyUI_Seedance.git into ComfyUI/custom_nodes, pip install -r requirements.txt (just requests>=2.28), restart, and add an API key from api.seedance.nz/console via the Seedance API Config node or the SEEDANCE_API_KEY env var.

The honest take

Wan's image quality was never really the point - it's a video model that happens to do good stills, and the community said so back when 2.1 was local. If you want a truly great text-to-image model you could just as easily run an open one locally for free, forever. This node earns its keep when you want the closed 2.7 model specifically, or when you want a reference-image editing model and don't want to fuss with local weights. Treat i2i-pro as the upcharge tier, test with a cheap T2I first, and remember every pixel is metered - leave thinking_mode off until you've seen whether the default path is good enough for your shot.