Wan Image to Image

Wan is the name you probably associate with open-source video - the Apache 2.0 family that a dozen local video models are built on. But Alibaba stopped releasing new Wan weights after 2.2, and the numbered series lives behind an API now. This node is that API's image-editing model, surfaced inside ComfyUI as a built-in partner node. At roughly 6.3 credits a run it's one of the cheapest paid-image nodes in the whole lineup, and it does something the local Wan stacks don't: two-image fusion.

What it's for

Two jobs, one node:

• Single-image editing - hand it an image and a prompt ("make it sunset", "turn the car red") and it rewrites toward your description.
• Multi-image fusion - feed it two images and it blends them. This is the genuinely unusual capability. One input image is editing; two inputs is "make a picture that contains both of these things." Most hosted editors won't do that without a whole reference-system rig.

The output is fixed at 1.6 megapixels, and its aspect ratio matches whatever you fed in - there's no width/height control, which is a deliberate simplification the node makes for you. The model dropdown has exactly one entry right now, wan2.5-i2i-preview, so don't go hunting for settings that aren't there.

Inputs that matter

• image - one or two images. That's the whole edit/fuse decision; the node enforces a max of 2.
• prompt - your instructions. Wan's image model handles English and Chinese prompts natively, so bilingual teams can just write in whatever's natural.
• negative_prompt - optional "what to avoid." Nice to have, and a rarity among the newer partner nodes, so use it.
• seed - reproducibility, though hosted models treat it as a suggestion. Fine for varying outputs.
• watermark - adds an AI-generated watermark marker. Off by default; turn it on if you want the provenance flag.

Output is a single IMAGE that flows straight into Preview/Save or into a Wan image-to-video node for the obvious next step.

Setup and the real talk

It ships with ComfyUI core - no install. You need a Comfy account with credits (Settings → User to log in, Settings → Credits to top up), and you should be on a recent ComfyUI since partner nodes iterate fast. The cost here is the selling point: about $0.03 per run, cheap enough that you can afford to batch a few variations instead of babysitting one perfect prompt. Where people get surprised: the 1.6MP ceiling is real, so don't expect to upscale-by-API here - if you want the output bigger, run it through your local upscaler afterward, which is free and usually better anyway. And remember this is a hosted service: your images leave your machine, so don't feed it anything you wouldn't upload to a cloud editor.