Skip to content
Playcut Docs

Nano Banana 2: aspect ratios, 4K and credit pricing

Nano Banana 2 is Google’s gemini-3.1-flash-image-preview — the fast, cheap sibling of Nano Banana Pro on the same Gemini image stack. Pick it when you are iterating, generating in volume, or batching variations: it costs 34 credits per 1K or 2K image and 43 credits at 4K, roughly half of Pro’s price for the same aspect ratios and the same resolution range. Pro remains the default in Playcut; Nano Banana 2 is an opt-in choice in the model picker or via the model field on the API.

Specs at a glance

Section titled “Specs at a glance”
Model idgemini-3.1-flash-image-preview
ProviderGoogle
ModesText-to-image, reference-to-image
Aspect ratios1:1 2:3 3:2 3:4 4:3 4:5 5:4 9:16 16:9 21:9
Resolutions1K, 2K, 4K (default 1K)
DurationN/A — still image
Max reference images16 accepted by Playcut; Google documents up to 10 object + 4 character + 3 style images (see Limitations)
Credits34 (1K), 34 (2K), 43 (4K) — reference-to-image costs the same

Aspect ratios and resolutions must both be valid for the model you select. Playcut checks the combination before credits are held, so an unsupported value returns an error instead of charging you.

Nano Banana 2 vs Nano Banana Pro

Section titled “Nano Banana 2 vs Nano Banana Pro”
Nano Banana 2Nano Banana Pro
Model idgemini-3.1-flash-image-previewgemini-3-pro-image-preview
Aspect ratiosSame 10Same 10
Resolutions1K / 2K / 4K1K / 2K / 4K
Modest2i, reference-to-imaget2i, reference-to-image, multi-region edit
Credits @ 1K / 2K3467
Credits @ 4K4384
Default in PlaycutNoYes

The only capability Pro has that Nano Banana 2 does not is multi-region edit. Everything else is a price/quality trade, not a feature gap.

Modes

Section titled “Modes”

Text-to-image

Section titled “Text-to-image”

Prompt only. Optional: aspectRatio, imageSize, negativePrompt, seed, outputFormat.

  • Tool: generate-image-from-text
  • Requires: prompt
  • Cost: 34 credits (1K, 2K) · 43 credits (4K)

Reference-to-image

Section titled “Reference-to-image”

Supply one or more existing Playcut assets as visual anchors, plus a prompt that directs the new composition (“put this product in a sunlit kitchen”).

  • Tool: generate-image-from-reference
  • Requires: prompt + referenceAssetIds (at least 1)
  • Cost: identical to text-to-image — 34 credits (1K, 2K) · 43 credits (4K)

References are sent to Google positionally as inline image parts. There is no named-slot syntax; order is the only signal about which reference matters most.

Multi-region edit — not available

Section titled “Multi-region edit — not available”

Playcut’s multi-region edit workflow runs on Nano Banana Pro only. There is no Nano Banana 2 path for it.

Pricing

Section titled “Pricing”
ResolutionPlaycut creditsGoogle’s published cost per image
51220$0.045
1K34$0.067
2K34$0.101
4K43$0.151
Reference-to-imageSame as aboveSame as above

Provider costs above are Google’s own published per-image prices for gemini-3.1-flash-image-preview, verified against the Gemini API pricing page on 2026-07-21. There is no separate surcharge for reference images — a reference-to-image call at 2K costs the same 34 credits as a text-to-image call at 2K.

Limitations & gotchas

Section titled “Limitations & gotchas”
  • No wide-strip aspect ratios. See the callout at the top. The supported list is exactly the 10 ratios in the specs table.
  • 512px is Flash-only. Sending imageSize: "512" to Nano Banana Pro or Grok returns a 400 before credits are held.
  • Reference cap is 14, matching Google’s composite limit. Google’s input-image table for this model documents up to 10 object images and 4 character images (14 total; style references are Pro-only — verified against ai.google.dev on 2026-07-23). Categories are guidance, not request slots — references are sent positionally. In practice 1–4 references produce the most predictable results.
  • The 14-reference cap is enforced server-side. referenceAssetIds is validated as 1–14 asset IDs on the REST tools endpoint and on MCP, and the API rejects a longer array with a 400 before any credits are held — not a silent truncation.
  • Failed references are silent. If a reference image cannot be loaded, generation proceeds text-only and the generation is still billed. Verify your referenceAssetIds resolve before submitting.
  • One image per request. Multiple candidates per call are not enabled on this model. Use outputFormat (2, 2x2, 3x3, 4x4) if you want several panels — but note they all render inside a single output image, not as separate assets.
  • Resolution strings are case-sensitive upstream. Use 512 / uppercase 1K / 2K / 4K. Playcut normalizes before sending, but the API field only accepts those four values.
  • Thinking cannot be disabled. Nano Banana 2 is faster than Pro, but it is not a “no-thinking” model — expect real latency, not instant returns.
  • Every output carries a SynthID watermark. Google applies it to all Gemini image output; it is not removable.
  • Model id carries a -preview suffix. Playcut sends gemini-3.1-flash-image-preview. Google’s own samples now use gemini-3.1-flash-image. If the preview alias is retired upstream, this model path breaks.

How to generate

Section titled “How to generate”

Set model to gemini-3.1-flash-image-preview on either image tool. Omit it and you get Nano Banana Pro (the default) at roughly 3x the credits.

Terminal window
curl -X POST https://api.playcut.ai/api/v1/tools/generate-image-from-text/execute \
  -H "Authorization: Bearer $PLAYCUT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "A ceramic espresso cup on a sunlit marble counter, soft shadows",
    "model": "gemini-3.1-flash-image-preview",
    "aspectRatio": "4:5",
    "imageSize": "2K"
  }'

Both tools return an async acknowledgement — taskId, status, estimatedCredits. Poll with get-task-status until COMPLETED, then fetch the signed URL (valid ~1 hour) with get-asset. Full flow in the Quickstart.

{
  "data": {
    "taskId": "65f3a1b2c3d4e5f600000002",
    "status": "PENDING",
    "estimatedCredits": 34
  }
}
Section titled “Related”