Wan Bailian CN I2I Config

Here's the awkward thing about Wan: the community's beloved open models stop at Wan 2.2. Everything after that - 2.5, 2.6, 2.7 - Alibaba kept API-only, served through its Bailian/DashScope cloud. So the "Wan" you can actually run locally is a different model from the "Wan" that this node talks to. WanBailianCNI2IConfig is the bridge to that newer, closed, Wan 2.7 image model, and it knows it's only reachable with a DashScope key.

It's an image-to-image config: it takes a prompt plus a source image, and outputs a MEDIAHUB_IMAGE_REQUEST for MediaHubGenerateImage to execute. (The pack also has a T2I sibling, but if you're on this page you want edits.)

Inputs that matter

• image - the source IMAGE, JPEG-encoded to base64 and shipped with the request.
• prompt - the edit you want.
• model - wan2.7-image-pro (the flagship) or wan2.7-image (the standard/cheaper lane). The pro variant is where the text-rendering and fine-detail improvements live.
• size - 1K, 2K, or 4K output. No 3K tier here (unlike Seedream), but the 4K ceiling is still on the table for high-res edit jobs.
• n - how many results per request, 1–4. Multiplies cost.
• watermark - off by default; on if you need the AIGC marker.
• seed - -1 = random; set a value to make the edit reproducible.

Output: request → MediaHubGenerateImage.

Why you'd reach for it

Two reasons. First, if you want the current Wan 2.7 image quality, this is the only door - there's no open checkpoint to match it. Second, Wan models built their reputation partly on text rendering and layout, and wan2.7-image-pro keeps that strength; it's a solid pick for edits involving readable text. If you've been running Wan 2.2 locally and want to A/B it against the newest closed model, this node makes that comparison trivial.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/vantang/ComfyUI-MediaHub.git
/path/to/ComfyUI/python -m pip install -r ComfyUI-MediaHub/requirements.txt

Restart ComfyUI; node under MediaHub/Provider/Wan Bailian CN/Image. Dependencies are requests/Pillow/numpy. You need an Alibaba Cloud Bailian account and a DASHSCOPE_API_KEY - same key the Wan/HappyHorse video nodes use.

Common gotchas

The biggest conceptual trap is thinking this replaces your local Wan. It doesn't - it's a different, newer, closed model reached over the network, and it bills per request. The JPEG round-trip on your source image is the other thing to keep in mind: clean inputs produce cleaner edits, same as every I2I node in the pack. And n sneaks up on you - at 4K with n=4, you're not making one image, you're making four at a premium tier, so treat that combo like the price tag it is.