Nano Banana Pro: aspect ratios, 4K and pricing

Nano Banana Pro is Google’s flagship Gemini 3 image model, exposed on Playcut as gemini-3-pro-image-preview. It is the default image model — if you omit model on any image call, you get this one. Pick it when you need embedded text rendering (posters, packaging, signage, UI mockups) or the highest output quality. Nano Banana 2 also reaches 4K for about a third of the credits; Grok Imagine Image is the only image model that stops at 2K. Nano Banana Pro costs 67 credits at 1K or 2K and 84 credits at 4K per image.

Specs at a glance

Model id	gemini-3-pro-image-preview
Provider	Google
Modes	text-to-image, reference-to-image, multi-region edit
Aspect ratios	10 (see below)
Resolutions	1K, 2K, 4K (default 1K)
Duration	N/A — still image
Max reference images	16 on Playcut · Google documents up to 6 object images for this model
Credits	67 (1K/2K) · 84 (4K) · 25 flat (multi-region edit)

Aspect ratios

Ten ratios, and only these ten. They are identical for text-to-image and reference-to-image.

1:1  2:3  3:2  3:4  4:3  4:5  5:4  9:16  16:9  21:9

No ultrawide

Wide-strip ratios (1:4, 1:8, 4:1, 8:1) are not supported. They do not appear in Google’s documentation and an 8:1 request fails with a 400 at the provider. Playcut rejects them before credits are held.

Pixel dimensions per ratio

Ratio	1K	2K	4K
1:1	1024 × 1024	2048 × 2048	4096 × 4096
2:3	848 × 1264	1696 × 2528	3392 × 5056
3:2	1264 × 848	2528 × 1696	5056 × 3392
3:4	896 × 1200	1792 × 2400	3584 × 4800
4:3	1200 × 896	2400 × 1792	4800 × 3584
4:5	928 × 1152	1856 × 2304	3712 × 4608
5:4	1152 × 928	2304 × 1856	4608 × 3712
9:16	768 × 1376	1536 × 2752	3072 × 5504
16:9	1376 × 768	2752 × 1536	5504 × 3072
21:9	1584 × 672	3168 × 1344	6336 × 2688

Note

Google publishes no per-ratio dimension table. Only the square values (1024 / 2048 / 4096) are corroborated by Google’s own pricing page — the rest of this grid is Playcut’s table, used by our validator. Treat non-square dimensions as expected, not guaranteed.

Nano Banana Pro does not get the 512px (0.5K) tier. That tier is exclusive to Nano Banana 2, and Playcut does not sell it on either model.

Modes

Text to image

Prompt only. Choose an aspect ratio and a resolution; everything else is optional. This is the mode that runs when you call generate-image-from-text without a model.

Input	Required	Notes
prompt	yes	Plain text
aspectRatio	no	One of the 10 above; default 1:1
imageSize	no	1K, 2K, 4K; default 1K. Must be uppercase K

Reference to image

Same model, same ratios, same resolutions, same price. You supply reference images alongside the prompt and the model composes from them.

Input	Required	Notes
prompt	yes	Plain text
referenceAssetIds	yes	Playcut asset IDs you own — 1 to 16
aspectRatio / imageSize	no	Identical options to text-to-image

References are sent to Google as inline image parts appended after the text part. There is no named-slot syntax — you cannot write @ref1 in the prompt and expect the model to bind it to a specific image. The model sees the images positionally.

Playcut’s cap is 14 reference images, matching Google’s documented composite limit — a request with more (via API or MCP) is rejected before any credits are held. Google’s input-image table for Gemini 3 Pro Image breaks the 14 down as up to 6 object images, 5 character images, and 3 style-reference images (verified against ai.google.dev on 2026-07-23). These categories are guidance, not request slots — references are sent positionally and the model infers their role. If you need character or style references, Nano Banana 2 is the model Google documents them for.

Multi-region edit

Nano Banana Pro is the only model wired to Playcut’s multi-region edit surface: you mark regions on an existing image and describe a change per region. It bills a flat 25 credits per run regardless of how many regions you edit.

Multi-region edit is a Studio (web app) feature. It is not part of the public tools registry, so there is no /api/v1/tools/... endpoint for it today.

Pricing

All prices are per generated image, in Playcut credits.

Mode / tier	Credits
Text-to-image @ 1K	67
Text-to-image @ 2K	67
Text-to-image @ 4K	84
Reference-to-image	Same as text-to-image at the same tier
Multi-region edit	25 flat

1K and 2K cost the same — there is no reason to generate at 1K unless you want smaller files. 4K is a 25% credit premium over 2K.

For how this compares against the rest of the catalog, see Resolutions & pricing.

Provider cost

Google’s published cost for this model is $0.134 per image at 1K/2K and $0.24 at 4K (verified against Google’s pricing page on 2026-07-21). Playcut credit prices are fixed per tier and do not track provider price changes automatically.

Limitations & gotchas

• One image per request. Google rejects candidateCount > 1 on this model. Grid and storyboard layouts are produced as a single image via prompt composition, never as multiple candidates.
• Lowercase size strings are rejected by Google. 1k errors; 1K works. Playcut upper-cases the value for you before dispatch, but do not rely on that if you call Google directly.
• It is slow, by design. Thinking is always on for Gemini 3 image models and cannot be disabled. Nano Banana Pro is materially slower than Nano Banana 2 — that is the model, not the network.
• Every output carries a SynthID watermark. Google embeds it in all generated images. Playcut does not strip it.
• Reference failures are silent. If every reference image fails to load, the generation proceeds text-only and you are charged the full price for an image that ignored your references. Verify your asset IDs before submitting.
• Reference assets you do not own are skipped, not rejected — same failure shape as above.
• The 14-reference cap is enforced server-side. referenceAssetIds is validated as 1–14 asset IDs on both the REST tools endpoint and MCP, and the API rejects a longer array with a 400 before any credits are held — not a silent truncation.
• Image segmentation is not supported on Gemini 3 Pro. Pixel-mask workloads are not available here.
• Credits are held before dispatch. An invalid combination that reaches the provider costs you the hold and returns a 400. Playcut pre-validates aspect ratio and resolution per model to prevent this — stay inside the tables above.
• The model id carries a -preview suffix that Google’s current samples no longer use (gemini-3-pro-image). If Google retires the preview alias, this id changes.

How to generate

Every generation goes through the tools registry. See the Quickstart for session creation and asset download.

Text to image

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 '{
    "workerId": "65f3a1b2c3d4e5f600000001",
    "prompt": "A vintage travel poster for Kyoto in autumn, bold serif title text",
    "model": "gemini-3-pro-image-preview",
    "aspectRatio": "2:3",
    "imageSize": "4K"
  }'

Reference to image

curl -X POST https://api.playcut.ai/api/v1/tools/generate-image-from-reference/execute \
  -H "Authorization: Bearer $PLAYCUT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "workerId": "65f3a1b2c3d4e5f600000001",
    "prompt": "Put this product on a marble kitchen counter, morning light",
    "model": "gemini-3-pro-image-preview",
    "referenceAssetIds": ["65f3a1b2c3d4e5f600000003"],
    "aspectRatio": "4:5",
    "imageSize": "2K"
  }'

MCP

From Claude Desktop, Claude Code, or Cursor, the same two tools are exposed as MCP tools:

generate-image-from-text
generate-image-from-reference

Pass model: "gemini-3-pro-image-preview" in the tool input. See the MCP install guide.

Both tools return an async acknowledgement with a taskId. Poll get-task-status until it reports COMPLETED, then call get-asset with the resulting asset ID for a signed URL (valid ~1 hour). Image generations typically complete in seconds rather than minutes, and 4K takes longer than 1K/2K — Playcut publishes no latency SLA.

Related

All models The full Playcut catalog — three image models, seven video models, music and voice.

Nano Banana 2 Same 10 ratios, same resolutions, roughly a third of the credits — and faster.

Grok Imagine (Image) Cheapest image model, 14 aspect ratios including ultrawide, no 4K.

Aspect ratios by model Which ratio works on which model, across every image and video model.

Resolutions & pricing Credit cost per resolution tier across the whole catalog.