Qwen Image 2.1 Pro · Edit

Pro here means two specific upgrades over the plain 2.1 edit: every run is a 2K render, and one request can return up to six variants of the same prompt. The rest - ten reference images, no negative prompt, the rewrite machinery - is the family's shared shape.

About 5.4 credits ($0.054) per image, and the per-image part is the important bit: num_outputs 6 is six charges, not one.

The most-missed interaction in the pack

Two widgets sit next to each other and one gates the other:

• enable_prompt_expansion - default False. The service rewrites your brief into something fuller before generating.
• enable_thinking - default True. The author's tooltip: "Spend extra reasoning on the rewrite for better images, at the cost of a longer wait. Only works while prompt expansion is on."

So out of the box, thinking is on and doing nothing, because the rewrite it thinks about is switched off. If you want the smarter rewrite, turn expansion on first, then decide whether you want to wait for the reasoning pass. If you want your words followed literally - which on an edit is usually the correct instinct - leave both alone.

That's not a bug, it's a schema quirk faithfully mirrored into widgets, and it's the kind of thing you'd never guess from the node alone.

Fields worth setting

prompt - the tooltip carries an edit-specific warning: there is no negative prompt on this model, so "write what to leave out directly". "Keep the background, no extra people" is a legitimate instruction, not a wasted sentence.

image_urls - Autogrow, one to ten, ordered. The prompt tooltip asks you to refer to them as "Image 1, Image 2 and so on" - use that naming, this model reads it. Each reference raises the quote.

num_outputs - 1 to 6, default 1. Set it to 6 and you get six takes on the same instruction in one round trip; billed per image returned. Genuinely useful when you're hunting a composition and don't want to babysit six queue clicks.

resolution - 2K default, and per the tooltip "every tier costs the same". There is no cheaper tiny-draft mode on this node, which changes how you should explore: use the turbo sibling for cheap probing, then come back here.

aspect_ratio - model default by default on the edit node, i.e. omitted from the request so the size is inferred from the source images. Better default than the plain 2.1 edit's 1:1, honestly.

seed - fixed, so an accidental re-queue costs nothing.

Outputs: images (a batch, one per requested output) and task id. No alpha output on Pro - for transparency, use the plain Qwen Image 2.1 · Edit.

Install

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

Restart. Or install from ComfyUI Manager by searching PopNow. There's no dependency step - the pack declares none, and it uses only the aiohttp, torch, Pillow, NumPy and PyAV that ComfyUI already bundles. Python 3.10+ and a V3-API ComfyUI are the requirements that actually bite.

Then popnow.ai/api/console/keys → Settings → PopNow → restart twice if you want to be certain. The catalog is read at startup; a new model won't appear mid-session.

Honest caveats

The whole-frame re-emission problem is inherent to this editing family: pixels you didn't discuss move. At 2K it's less visible per-pixel but the drift is still there, and on a chain of edits faces wander. One edit, then composite.

Six outputs arrive as one IMAGE batch, so a Save Image node writes six files. That's convenient right up until you forget and wonder where the disk went.

Money, again: the quote is taken before submission and the node prints the final charge when it finishes, so you're never guessing. 40201 is an empty wallet, 40004 an unservable combination (it names the field), 40310 an unverified email. Refused or failed runs are refunded. Settings → PopNow → Max credits per run is the seatbelt - with num_outputs at 6 it's the difference between a test and a purchase.