Nano Banana 2 edit chain cost on Sume: 1K, 2K and 4K per pass
A Nano Banana 2 edit costs $0.10 at 1K, $0.15 at 2K and $0.20 at 4K on Sume. Cost of 3-, 5- and 10-pass chains, and when to draft at 1K and finish at 4K.

A Nano Banana 2 edit on Sume costs $0.10 at 1K, $0.15 at 2K and $0.20 at 4K, so a 5-pass chain is $0.50, $0.75 or $1.00 depending on the tier you run it at. The price follows the resolution you request, not the number of references or the prompt length.
Those are Sume's billed prices, which are the provider list price times 1.25, taken from the Sume Image API docs and catalog (read 2026-10-05). The table below turns them into chains, and the section after it shows why the cheapest chain is usually not all-1K.
What does each pass cost?
Every pass of a chain is a full POST /v1/images call with model: "google/nano-banana-2", so each one is billed. A chain of N passes costs N times the per-pass price, with no discount for repeated edits of the same picture.
| Chain length | All at 1K | All at 2K | All at 4K |
|---|---|---|---|
| 1 pass | $0.10 | $0.15 | $0.20 |
| 3 passes | $0.30 | $0.45 | $0.60 |
| 5 passes | $0.50 | $0.75 | $1.00 |
| 10 passes | $1.00 | $1.50 | $2.00 |
Should I draft at 1K and finish at 4K?
Resolution is the one price lever you control per pass, so use it. Run the exploratory passes, the ones where you are still deciding what to change, at 1K, and request 4K only for the last pass. A 5-pass chain with four drafts at 1K and one finish at 4K costs 4 x $0.10 + $0.20 = $0.60, which is $0.40 less than five passes at 4K and only $0.10 more than five at 1K.
The catch is that the last pass then re-renders the picture at a new size, so the finish should be a fresh edit of the last approved draft with the same prompt, not a separate upscale step. Save each pass as a PNG before feeding it back; why PNG between passes explains how JPEG loss compounds.
How does a chain compare with a single edit and a re-roll?
Chains are priced per call, so the question is how many calls the picture needs. A model that gets a change right on the first try at 1K is $0.10; a model that needs three tries is $0.30 whatever its sticker price. Count calls per approved image from your own log, not from a list price.
The same arithmetic applies to every model on Sume. The Ideogram 4.5 ten-edit chain post prices that model's chain the same way, and the numbers are a different shape because Ideogram's price follows quality and not size.
- Pick the resolution per pass, not per project:
resolutionis a request field. - Keep one
Idempotency-Keyper pass so a retry returns the original job instead of billing again. - Read
usage.costfrom each response and sum it. That is the billed amount, and it is the number to trust over this table.
What does a reference-heavy chain add?
Nothing, according to the pricing in the docs: the table has one price per resolution and no input-image fee. Nano Banana 2 accepts up to 10 input references, so you can send the source plus a logo and a style sheet in each pass without changing the price. See reference limits by model for the full list. Input size can still change how long a call takes, and slow 4K calls fall back to a 202 job envelope after the 30 second wait, so branch on the status code and poll status_url rather than treating 202 as a failure.
A budget rule you can keep
Set a ceiling in dollars per approved image, not per call. If you allow $0.60 for a finished picture, that is six 1K drafts and no finish, or four drafts plus a 4K finish, or three 2K passes plus a 4K finish with a few cents to spare. Write the rule down once and let the script stop at it.
One caveat to keep in view: nothing on this page says one tier gives a better edit than another for your picture. The docs publish prices and limits, not quality, so the only way to know whether 1K drafts predict the 4K result is to try ten of your own images at both and compare them.
Sources
Related posts
More in Pricing
- Nano Banana 2 thumbnail tier: Google 0.5K to 4K vs Sume billed
Google's 2026 Nano Banana 2 per-image prices from 0.5K to 4K next to what Sume bills, to pick a thumbnail tier by cost rather than habit.
- Nightly voice-agent test calls: 200 utterances, a month of TTS
A nightly regression run of 200 utterances at 100 characters is 600,000 characters a month: $9.00 on MAI Flash, $13.20 on MAI-Voice-2.1, $28.50 on Sume.
- 360p drafts before a 1080p final: when Omni drafting pays off
A 10 s Omni 360p draft costs $0.38 and a 1080p take $1.88. From the second attempt on, drafting first is cheaper. Break-even table on Sume, read 2026-10-05.
- Omni 4K costs twice 1080p per second on Sume: is it worth it?
On Sume, Omni 4K is $0.375 a second against $0.1875 at 1080p: exactly double. A 10 s clip is $3.75 versus $1.88. When the extra pixels pay for themselves.
Written by Sume