Nano Banana Pro · Text to Image

The use case, stated plainly

Ask a local checkpoint for a three-column infographic with a headline, four labelled icons and a footnote, and what you get back is a poster-shaped lie - beautiful text-like noise. Ask Nano Banana Pro and you get the infographic. That's the trade: about 16.08 credits an image for a model that plans a layout before it commits pixels, renders text in multiple languages, and outputs natively at 4K.

Google's flagship image model is also the most opinionated about what it won't draw, and the most famous for being technically excellent and a bit safe. The community read on the Gemini image line has long been "powerful, unimaginative" - treat it as a brilliant layout engine rather than your wildest collaborator. If you want weird, run it locally; if you want a menu with correct prices on it, come here.

How the node runs

Plain API-node mechanics: your prompt and settings go to PopNow, the run gets a quote, the task is submitted with an idempotency key so a retry can't double-charge, and the node polls until it settles before downloading the results into a normal IMAGE batch. No weights, no sampler, no VRAM. The pack declares no dependencies at all because everything it needs already ships with ComfyUI.

Two consequences you can't engineer away. Your prompt leaves the machine. And the model's filter is the model's - there's no local patch, because there are no local weights, which is the same line every API node in this ecosystem runs into.

Inputs

• prompt - describe the image. For this model, describe the structure too: the hierarchy, how many text blocks, what each says, where things sit. It rewards being written to like a designer's brief rather than a Danbooru tag string, and quoting the literal words you want rendered is the single highest-value habit.
• aspect_ratio - 16:9 by default. Posters, slides and infographics usually want a deliberate choice here rather than the default.
• resolution - the output size tier, 1k default. The top tier is where the native-4K advantage actually shows; at 1k you're buying the layout reasoning, not the pixels.

That's the whole node. No search toggles, no output-format widget - Pro gets the essentials and nothing else.

Outputs: images (IMAGE, batched when several come back) and task id (STRING), which is your handle in the API console and for support.

Install

comfy node install popnow-comfyui

or Manager → search PopNow → Install. Restart ComfyUI, then add a key: create one at popnow.ai/api/console/keys and paste it in Settings > PopNow, or set POPNOW_API_KEY on a server. Restart once more, because the node catalog is fetched from your account at start and not bundled. If a stale custom_nodes/popnow_api.py is lying around from the pre-pack single-file node, delete it.

Nothing downloads. You need a reasonably recent ComfyUI - the pack is built on the V3 node API - and that's the entire prerequisite list.

Budget and breakage

At ~16 credits an image this is the priciest image node in the pack, so do the boring work first: write the prompt against the cheap 5-credit Nano Banana 2 node until the layout is right, then run the keeper here at the tier you intend to ship. Settings > PopNow > Max credits per run exists for exactly this and blocks a run whose quote exceeds your ceiling before any credits are held.

Codes: PopNow 401 bad key, 40201 out of credits, 400 field-level rejection naming the culprit, 40004 an unservable combination, 40310 unverified email, content_rejected refunded. Errors trailing a request_id are support tickets in disguise - copy the id.

No seed control, so every queue is a new paid image; and cancel doesn't stop the job, it stops your wait, with the task still billed if it lands.

One caution that applies to this whole pack: PopNow has no meaningful community footprint - I couldn't find a single discussion thread about it - so nobody has published the edge cases for you. It's MIT and the source is short; keep the key in the environment, start with a small credit pack, and don't wire a client deliverable to a vendor you haven't stress-tested.