67 credits an image.
And when it is worth it.
Nano Banana Pro is Google's flagship Gemini 3 image model and Playcut's default. Change nothing in the picker and this is what renders: 67 credits at 1K and 2K, 84 credits at 4K, held against your balance the moment you submit.
This page is the buying decision, not the spec sheet — what nano banana pro pricing looks like per image, where the flat 25-credit multi-region edit fits, and the nine limits worth knowing before you spend.
Rates in credits · you pick the model, every time
What Nano Banana Pro costs on Playcut
Four numbers cover every way you can spend on this model. Text-to-image and reference-to-image cost exactly the same here, so attaching reference images adds nothing to the bill.
Every rate is in Playcut credits — what your balance is debited, not a provider's list price. The third column converts that into finished runs on a Pro plan's monthly allowance; the pricing page covers the other tiers.
1K and 2K sit on the same credit baseline before the model multiplier, so generating small saves file size and download time, never credits. If you are choosing between them on budget, there is no budget difference — pick 2K. 4K is the one tier that costs more, at 84 credits, and accounts without a paid subscription are capped at 2K.
Multi-region edit — flat 25 credits
Open a finished image in the editor, draw over up to eight areas, and give each area its own instruction. The run costs a flat 25 credits however many regions you mark, because the worker still makes a single call.
It always runs on Nano Banana Pro — the request body carries no model field, so there is nothing to switch. It is also the one Nano Banana Pro flow that never leaves the studio: no MCP tool, no public REST tool, web editor only.
Nano Banana Pro against Nano Banana 2, tier by tier
The same table for Google's cheaper image model. Every tier the two share costs roughly half on Nano Banana 2. Only one row belongs to Nano Banana 2 alone — the 512px tier — while Pro holds multi-region edit and the style-reference kind.
Nine limits before you spend
Every one of these is enforced in our own code or the provider's. They are the things that change what a project costs, framed as budget rather than as reference material.
- A dead reference id costs you a full render
- Bad asset ids do not stop the job. The worker logs a warning, drops that reference and renders anyway — text-only if every one of them fails. You pay 67 credits for an image that ignored the brief. A ten-shot set built off a stale id list burns 670 credits before anybody notices.
- Credits leave your balance at submit, not at delivery
- The hold lands the moment the task is dispatched. Model, aspect ratio, resolution and reference count are pre-validated, so a malformed request 400s before any credits move. If the provider refuses the prompt after that, the task goes FAILED and the full hold is reversed. A refusal costs you the round trip, not the 67 credits.
- One image per call — and one field that can quietly quadruple the hold
- Every Pro call returns a single image; grids of 2, 4, 9 or 16 panels are composition inside that one image. The studio and both API tools submit a count of one, so the hold matches what you receive. The web app's own image endpoint also accepts a numberOfImages field up to four, which multiplies the hold without multiplying the output. Leave it alone.
- There is no cheap thumbnail tier on Pro
- The 20-credit 512px tier belongs to Nano Banana 2 alone. Ask Pro for it and the request 400s before the hold — no silent promotion to 1K, no surprise 67-credit bill. Moodboard and thumbnail passes mean switching models, which is a picker change rather than a plan change.
- It is the most expensive image model on the platform
- 67 credits against 34 on Nano Banana 2 and 10 on Grok Imagine Image at the same 1K tier — about 2 Nano Banana 2 renders per Pro render, or 6 Grok renders. Being the default protects you from none of that. It is the default because our own picker rates it highest on quality, and that rating is an in-house preview score rather than a published benchmark.
- Rendering at 1K saves you nothing
- Both tiers sit on the same credit baseline before the model multiplier, so the smaller file is free bytes, never free credits. Pick 2K unless something downstream needs the lighter asset. Only 4K moves the number, at 84 credits.
- No ultrawide strips, on either Google model
- Ten ratios is the whole list, and 21:9 is as wide as it goes. Cinema-strip requests like 4:1 or 8:1 do not exist upstream and are rejected before dispatch. If a banner slot needs 4:1, budget for a crop out of 21:9 rather than a native render.
- 4K needs a paid plan
- Accounts without a subscription are capped at 2K, so the 84-credit render is unavailable until a plan is attached. The request comes back as a 403 with an upgrade message rather than silently downgrading the resolution you asked for. Nothing is billed for the attempt.
- Multi-region edit cannot be scripted
- No MCP tool, no REST tool — the web editor is the only route in, so it cannot sit inside an automated pipeline. Eight marked regions per run is a hard ceiling, enforced on the request body. At a flat 25 credits a run it undercuts a 67-credit Pro re-render, which is the trade for clicking through it by hand.
Nano Banana Pro vs Nano Banana 2
Nano Banana 2 is the other Google image model on Playcut, and the gap between them is narrower than the price implies. Both take the same ten aspect ratios. Both accept up to 14 reference images. Both reach 4K.
What differs is credits and one resolution tier. Nano Banana 2 costs 20 credits at 512px, 34 at 1K and 2K, and 43 at 4K — roughly half of Pro at every tier they share. The 512px budget tier is exclusive to Nano Banana 2, and Pro rejects that size upstream, so our API returns a 400 before any credits move.
Our catalog also marks them differently. Nano Banana Pro is stable; Nano Banana 2 is preview. The studio picker rates Pro higher on quality and lower on speed, but those are Playcut's own preview ratings rather than a published benchmark, so treat them as a starting opinion.
When Nano Banana 2 is the right call
Treat Nano Banana 2 as the iteration model and Pro as the delivery model. The Pro plan's 2,000 monthly credits buy about 29 Nano Banana Pro renders at 1K; the same allowance buys about 58 on Nano Banana 2. If you expect to regenerate a brief four or five times before you like it, drafting cheap and re-running only the winners on Pro reaches the same final frame for less.
Take a twelve-frame product launch. Draft all twelve on Nano Banana 2 at 34 credits each (408 total), promote the four keepers to Pro at 2K (268), then fix two of those with multi-region edit instead of re-rendering (50). The set lands at 726 credits. Running the same twelve entirely on Pro with two re-renders costs 938 — 212 credits more for the same delivered frames.
The reverse case is just as clear. When the image is going to a client, onto packaging or into paid media, the extra 33 credits are a rounding error against the cost of getting it wrong. For a delivery frame, the default is the safe pick rather than the frugal one.
What 67 credits actually buys
One image, and one image is the unit you budget in. Nano Banana rejects a candidate count above one, so there is no cheap way to see four options at once — four options is four holds. Decide your iteration count before you decide your resolution.
Grid output is the one place that unit works for you. Ask for a sixteen-panel contact sheet and all sixteen panels render inside the single billed image, so it costs 67 credits at 1K rather than 1,072. The catch is that one weak panel re-runs the whole sheet at full price.
The controls you get for the price: ten aspect ratios, three resolution tiers at 1K, 2K and 4K, and up to 14 reference images. Left alone, a generation defaults to 1:1 at 1K.
The ratio list itself, the pixel dimensions behind each one, error semantics and request bodies live in the full Nano Banana Pro spec sheet in our developer documentation. This page stops at the buying decision on purpose.
Reference images
The 14-reference cap is Google's composite limit, and Nano Banana 2 carries the same 14. The two models are tied on count and differ only in the kinds Google documents: Pro splits its 14 across objects, characters and style references, Nano Banana 2 across objects and characters. Grok Imagine Image, the third image model here, stops at two.
Send more than 14 and you get a 400 before any credits are held — no silent truncation, and no half-briefed render you have to pay for. That check sits alongside the model, ratio and resolution validation, which is why over-stuffing the array is one of the few mistakes here that costs nothing.
How to run it
In the studio, doing nothing selects it. Nano Banana Pro is the default image model, so a fresh workspace renders on it without touching a setting. Model, resolution, aspect ratio and reference images are all chosen per generation — Playcut does not route your prompt to a model, and nothing is swapped behind your back.
Over MCP, two tools reach it: generate-image-from-text and generate-image-from-reference. Both are public and both default to Pro. Both also cap the prompt at 8,000 characters, a limit that exists on the tools API and not on the studio form. They return a task id rather than a file, so an agent polls for the result instead of receiving it inline, and neither tool exposes Grok Imagine.
Over REST the shape is the same: POST /api/v1/tools/{name}/execute with an API key, same tools, same credit balance. Field names and request bodies are in the docs; setup and the full tool list are on the MCP and REST API page.
What a month of credits buys
At 67 credits a render, Hobby's 500 monthly credits are about 7 Nano Banana Pro images, Pro's 2,000 about 29, Studio's 6,000 about 89, and Agency's 10,000 per seat about 149. Multi-region edits at 25 credits stretch far further — 80 of them on the Pro plan.
That arithmetic is the honest argument for mixing models rather than defaulting through a whole campaign. Draft on Nano Banana 2 or Grok Imagine, promote the survivors to Pro, and edit rather than re-render when a finished frame needs a fix.
Everything lands on one balance in one workspace, so switching models mid-project costs nothing but credits. The wider set of trade-offs sits on the model comparison hub, the production flows on the AI image generator page, and plan allowances on the pricing page.
Common questions
How much does Nano Banana Pro cost on Playcut? +
67 credits per image at 1K or 2K, and 84 credits at 4K. Multi-region edit is a flat 25 credits per run, no matter how many regions you mark. The pricing page shows how many credits each plan includes.
Is Nano Banana Pro worth double the credits over Nano Banana 2? +
For delivery work, usually. For iteration, usually not. They share the ten aspect ratios, the 14-reference cap and the 4K ceiling, but Pro adds two things Nano Banana 2 has no route to: multi-region edit, and the style-reference kind in Google's input table. A common pattern is drafting at 34 credits and re-running only the winners at 67.
Why do 1K and 2K cost the same? +
Because the credit baseline is set per resolution tier, and 1K and 2K share that baseline before the model multiplier is applied. Generating at 1K saves file size and download time, never credits. If you were choosing between them for budget reasons, there is no budget difference — pick 2K.
Can I use Nano Banana Pro over MCP or the REST API? +
Yes, through two tools: generate-image-from-text and generate-image-from-reference. It is the default on both, so omitting the model parameter gets you Pro. Those two tools expose only the two Google image models — Grok Imagine is not selectable there — and multi-region edit has no tool at all, so it stays Studio-only.
How many reference images can Nano Banana Pro take? +
Up to 14, which is Google's composite cap. Nano Banana 2 takes 14 as well, so the two are tied on count and differ only in the kinds Google documents: Pro splits its 14 across objects, characters and style references, Nano Banana 2 across objects and characters. More than 14 is rejected with a 400 before any credits are held.
Does Playcut pick the image model for me? +
No. You pick, every time. Nano Banana Pro is the default, so if you never open the model picker in the studio that is what renders, at the price shown next to it. Model, resolution, aspect ratio and reference images are all set per generation, and nothing is swapped behind your back.
Keep reading
Building against the API? The full Nano Banana Pro spec sheet carries the pixel dimensions, parameter names and error codes this page deliberately leaves out.
Ready to make something cinematic?
Pick a plan. Hobby starts with a 7-day trial — cancel within 7 days at no charge.