Image model
Nano Banana 2 API and pricing — without the API key
SynthPulse runs Google's Nano Banana 2, Nano Banana Pro and the original Nano Banana for you. Generate and edit images in your browser at a flat per-image credit price — no key to provision, no per-request invoice, no quota to negotiate.
Nano Banana 2 pricing, per image
| Nano Banana 2 · 1K | 20 credits · about $0.13–$0.25 per image |
|---|---|
| Nano Banana 2 · 2K | 30 credits · about $0.20–$0.37 per image |
| Nano Banana Pro · 1K | 45 credits · about $0.30–$0.56 per image |
| Nano Banana (original) · 1K | 10 credits · about $0.07–$0.12 per image |
| What a credit costs | $0.0067–$0.0125 per credit, depending on which credit pack you buy |
| Reference images | Up to 10 per prompt on Nano Banana 2 and Nano Banana Pro |
| Modes | Text-to-image and image-to-image, charged the same |
| Billing | Charged per finished image. A generation that fails is refunded automatically. |
Where it earns its price, and where it doesn't
Good at
- A fixed price per image. 1K is 20 credits, and that number sits on the Generate button before you press it.
- Editing costs what generating costs. Attach up to ten reference images and the bill does not move.
- Nano Banana Pro charges one flat 45 credits at 1K and at 2K, so the resolution question only exists on the cheaper row.
- A generation that fails refunds itself. Nobody files a ticket over 20 credits.
Not for
- 1K and 2K are the only two rungs here. The model's own 4K tier is not wired into the picker.
- The original Nano Banana is pinned to 1K. The resolution switch does nothing on it, because the endpoint behind it has no resolution field at all.
- Quantity gets no discount. Four 1K images from one submission is 80 credits, four separate charges.
- There is no free allowance. The account is free; the first image is not.
- No REST API and no API key. Everything here happens in the browser or in batch mode, not in your build script.
Three models, one name
The image picker carries three Nano Banana rows and they are not close in price. The original costs 10 credits an image. Nano Banana 2 costs 20. Nano Banana Pro costs 45, which is 4.5x the original for the same single press of Generate.
They do not take the same settings either. Nano Banana 2 and Nano Banana Pro both accept 1K or 2K and up to ten reference images. The original accepts neither: it is pinned to 1K, and it has no entry in our per-model reference cap table, so the upload field lets sixteen through on a default nobody has checked against the model. Read that 16 as our fallback, not as a capability.
Every credit figure on this page is read from the same pricing config the Generate button reads. To put the other image models against the same numbers, they are all listed in the model directory.
Ten reference slots, and what they are for
Ten references per prompt is more room than most edit jobs need, and that is the useful part: one slot for the subject, one for the background plate, one for the lighting you want copied, one for the typeface on a label. What the model cannot do is work out which is which, so give each one a job in the prompt.
Text-to-image and image-to-image cost the same on all three rows, so starting from a frame you already like never costs extra. If you are mostly editing rather than generating from nothing, the image-to-image guide compares every model here that takes a reference image, not only these three.
What 2K costs, and on which row
On Nano Banana 2, 2K multiplies the base by 1.5: 20 credits becomes 30. Nano Banana Pro has no resolution multiplier at all, so it is 45 at 1K and 45 at 2K. The gap therefore depends on which resolution you compare at: 2.25x at 1K, 1.5x at 2K. If you already know the render has to be 2K, Pro is a much smaller step up than its headline price suggests.
Quantity gets no discount either. Four 1K images from one submission are four separate charges totalling 80 credits, and whichever one fails is the one refunded. What a credit itself costs depends on the pack you bought; the range is on the pricing page, and the table above has already turned it into a per-image figure.
Spending three tiers in the right order
The expensive row does not fix a bad composition; it renders the same composition with more detail. So the order that wastes least is dull: rough the framing out on the original at 10 credits, switch to Nano Banana 2 at 20 once the layout holds and the lettering has to be legible, and spend 45 on Pro for the one frame that ships. Six drafts, three refinements, one final render is 165 credits, against 450 for the same ten runs on Pro.
Only three settings actually move the bill: which model, resolution and quantity. The rest are free, direction included — image-to-image costs exactly what text-to-image costs. The walkthrough of the image workspace goes through all five controls in the order they matter, if you would rather watch it happen than reason about it.
Writing for it
Two habits do most of the work. Put the exact words you want rendered in quotation marks and say where on the image they sit; that is the difference between readable lettering and decorative squiggles. Then write the negatives: what must not change, what must not appear. On an edit those matter more than the description, because the risk is no longer a wrong subject, it is the model quietly redrawing the one you gave it.
The prompts below are edit instructions, deliberately short. For the long text-to-image kind, the portraits and product shots and posters you write from nothing, the prompt guide has sixteen worth copying and four ways they usually go wrong.
Five edit prompts
Short instructions for an image you already have, with at least one reference attached. Copy one and swap the nouns.
- Remove something
Remove the cable running across the right side of the table. Rebuild the wood grain underneath it. Change nothing else in the frame.
Say what fills the hole. Left unsaid, the nearest object usually gets stretched across the gap.
- Fix the lettering
Change the label text to read "BATCH 07". Same typeface, same size, same position. Leave every other pixel as it is.
Editing text is the one job where naming what must not move matters more than describing what should.
- Extend the frame
Extend this image to 16:9 by adding more floor below and more wall above. Do not add objects, people or text to the new area.
Outpainting invents whatever you left undescribed, so describing the empty part is the whole trick.
- Combine two references
Use image 1 for the product and image 2 for the lighting and colour grade. Keep image 1's label text unchanged.
Number the references in the order you attached them and give each one a job. An unassigned reference bleeds everywhere.
- Keep the same person
Same person as the reference: same face, same hair, same jacket. Now standing in a doorway, three-quarter view, looking left.
List the features that have to survive. "Same person" on its own is not specific enough to hold a face across shots.
How the three compare, and what sits next to them
Credit costs are read from the same pricing config the Generate button reads. Reference limits are the per-model caps the upload field actually enforces. A dash means the picker does not offer that resolution for that model.
| 1K | 2K | Reference images | |
|---|---|---|---|
| Nano Banana 2 | 20 credits | 30 credits | 10 max |
| Nano Banana Pro | 45 credits | 45 credits | 10 max |
| Nano Banana (original) | 10 credits | not offered | 16 max |
| GPT Image 2 | 15 credits | 26 credits | 10 max |
| Seedream 5.0 Pro | 18 credits | 36 credits | 10 max |
Nano Banana 2 FAQ
Is there a Nano Banana 2 API?
Not one you can call. SynthPulse is a web app, and the point is that there is no key to provision, no rate limit to negotiate and no per-request invoice to reconcile: you get the model at a fixed credit price per image, with prompt history, a library and automatic refunds already attached. If you wanted an API in order to run one prompt set across a hundred images, the image page has a batch mode that does that with no code.
Is Nano Banana 2 free?
Not on SynthPulse. A 1K image is 20 credits and a 2K image is 30. Signing up is free and needs no card, but there is no trial allowance at the moment, so you need a plan or a one-time credit pack before the first render. The cheapest way to look around is the original Nano Banana row at 10 credits an image.
How much does a Nano Banana 2 image cost?
20 credits at 1K and 30 credits at 2K. Nano Banana Pro is 45 credits and the original Nano Banana is 10. A credit costs between $0.0067 and $0.0125 depending on which pack you buy, so one 1K Nano Banana 2 image lands at roughly $0.13–$0.25. Image-to-image costs the same as text-to-image, and a generation that fails is refunded.
How do I use Nano Banana 2?
Open the image workspace, pick Nano Banana 2 from the model list, describe what you want, and optionally drop in up to ten reference images to edit from. Choose 1K or 2K, check the credit cost shown on the Generate button, and run it. Results appear beside the form as they finish and stay in your library.
What prompts work best with Nano Banana 2?
Write one paragraph that names the subject, the framing, the lighting and the medium, then say what you do not want. Nano Banana 2 prompts reward specificity over keyword stacking: "a 50mm portrait of a ceramicist in a dusty workshop, late afternoon window light, shallow depth of field" beats a comma-separated pile of adjectives. When you need text inside the image, put the exact words in quotes and say where they go.
Nano Banana, Nano Banana 2 or Nano Banana Pro?
Use the original Nano Banana at 10 credits for cheap drafts and quick variations. Use Nano Banana 2 at 20 credits as the default — it is the one with the better instruction-following and legible text. Use Nano Banana Pro at 45 credits for the final render, when detail has to hold up at full size.
Nano Banana 2 or GPT Image 2?
20 credits against 15 at 1K, 30 against 26 at 2K, and both cap you at ten reference images. GPT Image 2 is the cheaper of the two at either size, though not by enough to decide on. Pick by output: one render of the same prompt on each costs less than the time spent reading comparisons.