Try Pro FREE for 2 weeksGet Started
Start for free

gemini-3-pro-image-preview — Image Edit AI Model

Nano Banana generates and edits images through Gemini's multimodal reasoning, so it follows a complicated instruction the way a language model does rather than the way diffusion guesses. Needs a Google AI key.

Run gemini-3-pro-image-previewBuild with it in Studio

Also runs in Film Camera Filter · On-Model Shots (Garment Locked)

What it does

Nano Banana generates and edits images through Gemini's multimodal reasoning rather than diffusion alone, so it follows a long or literal instruction more closely - text inside the picture, a layout described in words, an edit that has to respect what is already in frame.

Text in, one image out, with an optional source image on the image port when you want an edit rather than a fresh generation. It runs on Google's side against your own key, and the aspect ratio you pick decides the output resolution.

When to use it

  • Images with legible text in them - a label, a sign, a title card
  • Editing a supplied image while leaving the rest of the frame alone
  • Infographic-style layouts described in words rather than drawn
  • Wide banners and tall strips at 4:1, 1:4, 8:1 or 1:8 on Nano Banana 2.1
  • High-volume drafts on the cheapest tier before committing to Pro

Details

  • CategoryImage Edit
  • Year2025
  • LicenseGoogle Cloud Terms

Compliance & Provenance

  • ProviderGoogle · GPAI
  • EU AI Act RiskLimited Risk
  • Art. 50 TransparencyRequired — AI-generated outputs are marked

Inputs & Outputs

TextInputstring

Text prompt for image generation or editing instruction

ImageInputimageoptional

Input image for editing

Image 2Inputimageoptional

Second input image, e.g. a style reference or another subject

Image 3Inputimageoptional

Third input image

ImageOutputimage

Generated or edited image

Generated TextOutputstring

Text response from the model

Parameters

modelstringdefault Nano Banana Prooptions: Nano Banana Pro, Nano Banana 2.1, Nano Banana 2, +1 more

Model variant

Which Nano Banana variant to call. 'Nano Banana Pro' (default) — higher quality, slightly slower. 'Nano Banana 2.1' — Flash-speed successor to Nano Banana 2 with better text rendering and the panoramic ratios 4:1 / 1:4 / 8:1 / 1:8. 'Nano Banana 2' — kept for saved flows and now runs Nano Banana 2.1 (Google retires 3.1 Flash Image on 2026-10-29). 'Nano Banana 2 Lite' — cheapest and fastest, for drafts and high-volume calls. Pick Pro unless latency or cost is critical.

promptstringdefault

System prompt / instruction

Optional system prompt or instruction that frames the image task. The text input pin carries the user-visible request; this parameter adds extra context like role, style, or constraints. Leave blank to use the text input pin alone.

aspect_ratiostringdefault Match inputoptions: Match input, 1:1, 16:9, 9:16, 4:3, 3:4, 4:5, 5:4, 3:2, 2:3, +5 more

Aspect ratio

Output aspect ratio. 'Match input' (default) keeps the shape of the image on the Image port — use it for edits, where a different shape makes the model re-frame the picture instead of editing it. Pick an explicit ratio to force one: '1:1' square, '16:9' / '9:16' wide/tall, '4:3' / '3:4' classic photo, '4:5' / '5:4' portrait/landscape for feeds and print, '3:2' / '2:3' camera frame, '21:9' cinematic, and the panoramic '4:1' / '1:4' / '8:1' / '1:8' on Nano Banana 2.1 only (or 'Nano Banana 2', which now runs it). The pixel size follows image_size.

image_sizestringdefault 1Koptions: 1K, 2K, 4K

Output resolution (1K/2K/4K)

Output resolution. '1K' (default) is about one megapixel. '2K' and '4K' return a high-resolution image directly, so no upscaler step is needed; they cost more output tokens on the provider bill. 'Nano Banana 2 Lite' supports 1K only.

max_tokensintegerdefault 2048range 1–8192advanced

Max text tokens (1-8192)

Maximum tokens for the optional text response Gemini emits alongside the image (e.g. caption, rationale). Default 2048. Lower to save cost when you only need the image.

temperaturefloatdefault 0.7range 0–2advanced

Sampling temperature (0-2, not used by 2.1)

Sampling randomness for the text + image generation. 0 = deterministic / repeatable. 0.7 (default) = balanced creativity. >1.0 = more divergent. Not sent for Nano Banana 2.1 (or 'Nano Banana 2', which now runs it): Google rejects any sampling setting on that model. Keep at default unless explicitly tuning style variance.

seedintegerdefault -1range -1–2147483647advanced

Random seed (-1 = random)

Seed for generation. -1 (default) randomizes. A fixed seed makes repeated runs more alike but does not guarantee the same image. To keep an exact result for a follow-up edit, pin this node's result instead of relying on the seed. Not sent for Nano Banana 2.1 (or 'Nano Banana 2', which now runs it): Google rejects a seed on that model.

Limitations

  • 'Nano Banana 2' now runs Nano Banana 2.1: Google retires the model behind the old label on 2026-10-29, so both labels give the same result.
  • Nano Banana 2.1 ignores the temperature and seed settings - Google rejects any sampling control on it.
  • Runs on your own Google AI API key. Without one configured in the workspace the node cannot run.

Tags

  • image-generation
  • image-editing
  • multimodal
  • text-to-image

Alternatives in Image/Video Models

  • GPT Image

    OpenAI GPT-Image text-to-image / edit. Needs API key.

  • Omni

    Google Gemini Omni: text/image/video in, video out, with conversational editing. Which input ports apply depends on Task — Text to Video: Text only. Image to Video: + Image 1. Reference to Video: + Image 1–3. Edit / Extend: + Video (aspect ratio and resolution are ignored). Unused ports are dropped even if wired. Needs API key.

  • VEO

    Text/image to video with native audio, now run by Gemini Omni 1.1 Flash (Google retires VEO 3.1 on 2026-10-22). Needs API key.

  • Video Analysis

    Gemini video analysis. Feed a video and get back timecoded highlight segments. Needs API key.

Resources

All AI models