Nano Banana Pro · Edit

When 16 credits an image is the cheap option

Nano Banana Pro is Google's top image model, and the thing that separates it isn't a resolution number - it's that it reasons about the layout before it draws. Fourteen reference images, 4K native output, multi-language text rendering, and the tendency to get infographics, diagrams, UI mockups and dense posters structurally right instead of plausibly wrong.

That's what you're paying 16.08 credits an image for. If you're editing a photo of a jumper, this is overkill and you should use Nano Banana 2 Edit at a third of the price. If you're producing a labelled diagram, a slide with six text blocks and a legend, or a menu where the € symbols and the crop of the photo both have to be correct - this node will save you an afternoon of local inpainting and re-prompting, and the afternoon is worth more than the credits.

Mechanism

Standard API-node shape: your source images are encoded (PNG, or JPEG at high quality above 10 MiB), uploaded, and cached for twelve hours on the hash so a re-prompt doesn't resend them. The run is quoted, the task submitted under an idempotency key, polled to completion, and the files downloaded and decoded into an IMAGE batch. No local weights, no VRAM, no dependency install - the pack's own dependency list is empty.

The consequence you should be clear-eyed about: the images go to Google through PopNow's API, and the model's own content policy decides what comes back. There's no local bypass, because there's nothing local. For design work that's a non-issue; for anything the filter dislikes it's the whole story.

The inputs - all four of them

• prompt - "say what to change and what must stay the same." With this model you can also describe structure: "keep the six labels, re-letter them in French, keep the arrows where they are." It's a reasoning pass, so use it like a brief rather than a tag list.
• image_urls - autogrowing sockets, one to fourteen, ordered. The prompts refer to them by position, so name them by position.
• aspect_ratio - auto by default, which keeps the source's shape. Change it only if you're deliberately reframing.
• resolution - output size tier, 1k default; the 4K tier is where the typography advantage cashes out.

Note what's missing compared with the 2.1 Edit node: no enable_web_search, no enable_image_search, no output_format. Pro Edit is the plain, expensive, careful one.

Outputs: images (IMAGE) and task id (STRING).

Install and config

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

Restart. Create a key at popnow.ai/api/console/keys, paste it into Settings > PopNow or export POPNOW_API_KEY, then restart again so the node list is rebuilt from your account's catalog. Delete the old popnow_api.py if it's still sitting in custom_nodes.

Because this is the most expensive node in this corner of the catalog, spend five minutes on the settings page: Max credits per run stops any run whose quote exceeds your ceiling before anything is held, which is the difference between an experiment and a surprise.

Getting burned, cheaply

The most common failure isn't an error, it's a drifting edit: the thing you wanted changed changed, and so did the face, the logo and the background. Fix the prompt, not the model - state the invariant explicitly ("the face, the fabric texture and the background stay untouched").

Error codes worth memorising: PopNow 401 key, 40201 credits, 400 the field the message names, 40004 an unservable combination, 40310 unverified email, content_rejected refunded. If an error carries a request_id, keep it - support can find the request exactly.

Two more things. Every run is a fresh bill; there's no seed widget here to hold a result still. And cancelling only stops the wait - the task keeps generating server-side and is charged if it succeeds. As for trusting the vendor: PopNow has essentially no discussion history anywhere I could find, so the sensible posture is a small credit pack, the key in the environment rather than a widget, and a read of the source before you build anything permanent on top.