Nano Banana 2 · Text to Image

The middle rung

Google's Gemini image line, as PopNow exposes it, is a ladder: Nano Banana 2 at about 5.4 credits an image, 2.1 at 7.2, Pro at 16.08. This is the bottom of the ladder, and that's not an insult - Nano Banana 2 (Gemini 3.1 Flash, February 2026) is the speed-quality compromise, and for a lot of work the compromise is invisible. Product shots, editorial illustration, mood boards, a background you'll blur anyway: all comfortably inside its range, at 4K if you want to pay for the tier.

Where it earns its price is text and instruction-following. If your prompt says "a chalkboard reading SOUP - £4.50", it renders that, and it renders it on the first try more often than a local checkpoint does. That's the honest state of the quality gap in 2026: mostly closed, still open on typography.

What the node does

It's a quote-submit-poll-download wrapper. Your prompt and settings are priced first, the task is created with an idempotency key so a retry can't produce two bills, the node polls it to completion, downloads the images and hands you an IMAGE batch. No local model, no weights, no VRAM. The pack's dependency list is empty because it rides on libraries ComfyUI already ships.

Also worth noting: this family watermarks. Google's images carry an invisible SynthID mark - irrelevant for a thumbnail, relevant if you're publishing somewhere that cares about provenance. And every prompt leaves your machine and passes a filter you can't patch around.

Inputs

• prompt - what to draw. Quote the words that must appear literally.
• aspect_ratio - 16:9 by default. Set it deliberately; you're paying per image, so an accidental landscape crop of a portrait idea is money spent on a mistake.
• resolution - size tier, 1k default. Higher tiers cost more.
• enable_web_search - the model checks the web for current facts before drawing. Off by default.
• enable_image_search - it looks for reference images first. Off by default.

Those two toggles are the fun part. Ask for something time-sensitive - a lineup, a stadium, a product that shipped last month - and switching one on turns a plausible hallucination into something defensible. Cost is latency, not credits, so far as the node's own price display goes.

Outputs: images (IMAGE - a batch when the model returns several) and task id (STRING).

Install

comfy node install popnow-comfyui

Manager users: search PopNow, Install, restart. Then key up - popnow.ai/api/console/keys into Settings > PopNow, or POPNOW_API_KEY in the environment for servers - and restart a second time, because the model list comes from your account at startup. Remove any old custom_nodes/popnow_api.py. Recent ComfyUI required; the pack uses the V3 node API.

Problems you'll actually hit

PopNow 401 - key wrong or revoked. No PopNow API key is set - you didn't restart after pasting. 40201 - out of credits; at 5.4 per image that's a short session if you're exploring at 4K. 400 - the message names the field that can't take that value. 40004 - that setting combination isn't servable, commonly resolution against aspect ratio. 40310 - verify your email. content_rejected or generation_failed - the model declined or failed, and the credits are refunded.

No seed widget here, so each queue is a fresh paid image. To explore cheaply: lock the prompt at 1k, generate a handful, then re-run the keeper at a higher tier. And don't bother hitting cancel to save money - the job keeps running on PopNow and is charged if it succeeds.

If you need a per-run ceiling, Settings > PopNow > Max credits per run does exactly that; a run quoted above it stops before anything is held. Given how new this pack and vendor are - no community chatter to consult, no third-party writeups - a ceiling is a reasonable thing to set on day one.