Wan 2.7 Pro · Edit

Most edit nodes are single-source: one picture in, one instruction, one result. This one is a compositor. The summary promises "large 4K canvases and nine-image composites", which is a different job - you're stacking references and telling the model how they relate, rather than nudging one image.

At 11.25 credits ($0.1125) an image it's two and a half times the base Wan 2.7 edit tier, and you're paying for size and reference depth rather than for four variations. If your work is one source and a small change, the cheaper node is the right one. If it's "put this jacket on that model in that room, and here's the art direction", keep reading.

And the usual caveat for the whole Wan 2.7 line: closed weights. Alibaba shipped nothing open past Wan 2.2, so this is API-only whether you like it or not.

How it works

The pack generates one node per model from that model's published input schema, so the widgets you see are the endpoint's actual fields. image_urls is an autogrow socket: each connection adds another reference slot, each image is uploaded when you queue, and the request goes out as an ordered list. The job is quoted before any credits are held, submitted under an idempotency key so a retry can't double-charge, then polled with progress printed on the node. The result comes back as a batched IMAGE.

The prompt addresses your sources positionally - the first reference, the second, and so on - so build the list in the order the sentence needs. Getting that order right is most of the skill here.

The inputs that matter

prompt - describe the image you want back: subject, composition, style, lighting. With multi-image input, say which reference supplies what. "The person from the first image, wearing the jacket from the second, in the room from the third" is the register this model responds to.

image_urls - your sources. The notable thing about this tier is how many it takes: nine-image composites is the headline claim, and while the practical limit is your patience and the quote, two or three well-chosen references usually beat nine lukewarm ones.

bounding_boxes - a JSON string of rectangles in pixels from the top-left of each reference image, up to two per image. That's how you get local edits without building a mask graph in ComfyUI; mark the region, and the model treats it as the working area.

color_palette - JSON, 3 to 10 colours with optional percentage shares. For anything client-facing this is the most underrated widget in the pack: it's a direct art-direction channel that most people never open.

aspect_ratio defaults to model default, which follows the proportions of the last source image. Note what isn't here: no resolution dropdown and no output-count box. This tier runs at its own size, one image per run.

Outputs: images and task id.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/popnow-ai/popnow-comfyui

or search PopNow in Manager and install from there. Restart, add a key from popnow.ai/api/console/keys under Settings > PopNow (or set POPNOW_API_KEY), restart once more. No model files, no dependencies past ComfyUI itself, no GPU needed - the node uploads your references and downloads the results.

Coming from the old single-file node? Delete custom_nodes/popnow_api.py.

Where people get burned

Two JSON fields mean two chances to paste a malformed array. bounding_boxes and color_palette are strings, not native widgets, so build them carefully or generate them upstream.

The other one is expectation-setting on cost. Nine references and a 4K canvas sounds like it should be a big number, and the quote will tell you before anything is held - but only if you actually look. Set Settings > PopNow > Max credits per run to a figure you're comfortable with and let it refuse anything above that.

Errors: 40201 is out of credits, 40004 means the model can't serve that combination and names the field, 400 is a rejected value with the field named. Refused or failed generations are refunded. Interrupting the queue stops the wait, not the job - it keeps running and settles on PopNow's side.