Home/Models/Google/Nano Banana Pro
G

Nano Banana Pro

輸入:$1.5616/M
輸出:$9.3696/M
Nano Banana Pro 是一个用于通用型辅助的 AI 模型，适用于以文本为中心的工作流。它适合使用指令式提示来生成、转换和分析具有可控结构的内容。典型用例包括聊天助手、文档摘要、知识问答（QA）和工作流自动化。公开的技术细节有限；其集成方式与常见的 AI 助手模式一致，例如结构化输出、检索增强型提示，以及工具或函数调用。
新
商用
Playground
概览
功能亮点
定价
API
版本

Basic features

  • Text → Image: full prompt-driven generation with strong prompt adherence.
  • Image → Image (edits): fine, targeted edits with maintained subject/character consistency across multiple edits.
  • Maximum output resolution: up to 4K (examples and supported exact pixel sizes depend on aspect ratio; the API exposes 1K/2K/4K presets)
  • Iterative planning & self-correction: an internal “multi-stage” pipeline that detects and corrects common visual mistakes (perspective, text, fine geometry).
  • Advanced in-image text rendering: clear, legible multi-language text (short captions to long paragraphs) suitable for posters, mockups, and infographics.
  • 5 characters and fidelity for up to 14 objects/reference images in a single workflow.
  • Watermarking / provenance: all generated images include a SynthID watermark; model embeds C2PA metadata for provenance in some product integrations.

Gemini 3 Pro Image versions & naming

  • gemini-3-pro-image-preview
  • gemini-3-pro-image

Technical details

Architecture

  • Lineage / backbone: Nano Banana Pro be built on Google’s evolving Gemini image stack — specifically the new Gemini 3 Pro Image / GEMPIX 2 architecture (a higher-capacity multimodal image+text framework). That is an evolution from Gemini 2.5 Flash Image (the original “nano-banana”) into a natively multimodal image model with expanded vision-language reasoning capabilities.
  • Model behavior: native multimodality (image + text + world knowledge), explicit pipelines for multi-image fusion, and an internal staged planner that refines outputs over multiple passes rather than producing a single static sample. Early reports indicate stronger geometric/optical reasoning (glass, refraction) versus prior versions.
  • Thinking / internal refinement: The model uses a visible “thinking” process internally to refine composition (the API documents this behavior and notes those internal steps are not charged as final image tokens).
  • Grounding & tools: Supports Search grounding (can incorporate web facts into diagram/infographic generation). It also supports system instructions for more deterministic control.

Key API parameters:

  • thinking_level (low / high) to trade latency vs reasoning depth;
  • media_resolution (low/medium/high) to control image OCR/detail reading tokens;
  • generationConfig.imageConfig to control aspect ratio/resolution in image outputs.

Image limits:

  • Input modalities supported: Text and images (the model does not accept audio or video as image-generation inputs).
  • Max images per prompt: 14 (for the Gemini 3 Pro Image preview).
  • Max image size (upload): 7 MB per input image.
  • Supported aspect ratios: 1:1, 3:2, 16:9, 9:16, 21:9, etc.

Output images / tokens: high limits, with 4K/4096px supported.

Benchmark performance

Short summary: public/early benchmarks so far are mostly qualitative / community-driven, but consistently report substantial improvements in resolution, artifact reduction, and physical fidelity versus the original nano-banana (Gemini 2.5 Flash Image). Specific named “challenges” have shown clear visual gains, but there are not yet (public) standardized numeric benchmark tables from Google comparing v1 → v2 across standard image-generation metrics.

  • Qualitative community tests: Cleaner edges, sharper micro-details, truer colors, and more faithful prompt adherence (fewer hallucinated props, more consistent characters). Popular informal tests include the so-called “Wine Glass Test” and “Glass Burger Challenge”, where GEMPIX2 (Nano Banana Pro) handles transparency and refraction markedly better than earlier builds.
  • Text handling: Nano Banana Pro shows visibly improved typography and text placement inside images (a persistent weakness for many image models). Community comparisons indicate fewer garbled rendered glyphs.
  • Throughput / UX: faster iteration speed and a UX that performs multi-stage refinement on the back end so users see more reliable first-pass results (reducing manual re-rolls).

Limitations & risks

  • Content filters & detection: Platforms integrating the model (e.g., Whisk/third-party apps) may enable strict celebrity or likeness detection and block certain outputs, which affects creative workflows that rely on realistic celebrity likenesses.
  • Hallucination / reasoning edge cases: while improved, the model can still produce physically unrealistic artifacts, especially with dense symbolic text inside images or highly technical diagrams — though NB2 appears to reduce these errors versus earlier versions.
  • Safety & misuse: generative image models can be used to create problematic or harmful content. Google applies constraints, content filters, and the SynthID watermark to help with provenance; nevertheless, misuse has occurred (high-profile controversy tied to a Nano Banana generated image in a politically sensitive setting).

How Nano Banana Pro stacks up vs other models

  • Nano Banana Pro (GEMPIX 2 / Gemini 3 Pro Image) — strong mobile integration, multi-image fusion, iterative self-correction, 2K native/4K upscaling, tightly integrated into Google apps (Search, Photos, Workspace/Gemini). Best for workflows that need reliable edits, continuity, and integration with Google services.
  • Midjourney — excels at stylized artistic outputs and community-driven prompt engineering; not typically targeted at photo-accurate multi-image fusion or deep multimodal editing pipelines.
  • Stable Diffusion / open weights — fully open, highly customizable, and hostable locally; ecosystem of checkpoints and fine-tuning is a decisive advantage for research and offline usage. Less “one-click” mobile integration and less consistent multi-image editing coherence out-of-the-box than Nano Banana Pro.
  • Seedream 4.0 (ByteDance) — recently positioned explicitly as a Nano Banana competitor, emphasizing ultra-fast rendering, 2K output, and support for many reference images (up to six). Positioned as a pro/creator alternative.

(These comparisons are high level; pick a winner by matching the tool to your workflow: openness/customizability → Stable Diffusion; stylized art → Midjourney; integrated, consistent mobile editing with aggressive iteration → Nano Banana Pro/ Gemini 3 Pro image family.)

Real-world use cases

  • Mobile photo editing & creative filters (Google Photos integrations — restyling, background fusion, portrait recomposition).
  • Marketing & ad assets — fast concept generation, consistent brand characters across multiple frames/angles.
  • Concept art & storyboarding — multi-image fusion helps keep character continuity across panels.
  • E-commerce / product mockups — generate consistent product shots in different contexts/lighting conditions.
  • Rapid prototyping for AR/VR assets — high quality 2K/4K outputs that can be upscaled for immersive uses.
  • How to accessl gemini-3-pro-image(Nano Banana Pro) API

Required Steps

  • Log in to cometapi.com. If you are not our user yet, please register first
  • Get the access credential API key of the interface. Click “Add Token” at the API token in the personal center, get the token key: sk-xxxxx and submit.
  • Get the url of this site: https://api.cometapi.com/

Use Method

  1. Select the “gemini-3-pro-image” endpoint to send the API request and set the request body. The request method and request body are obtained from our website API doc. Our website also provides Apifox test for your convenience.
  2. Replace <YOUR_API_KEY> with your actual CometAPI key from your account.
  3. Insert your question or request into the content field—this is what the model will respond to.
  4. . Process the API response to get the generated answer.

CometAPI provides a fully compatible REST API—for seamless migration. Key details :

  • Base URL: https://api.cometapi.com/v1beta/models/gemini-3-pro-image-preview:generateContent
  • Model Names: gemini-3-pro-image
  • Authentication: Bearer YOUR_CometAPI_API_KEY header
  • Content-Type: application/json .

常见问题

Can Gemini 3 Pro Image generate 4K resolution images?

Yes, Nano Banana Pro (Gemini 3 Pro Image) supports native output up to 4K resolution with aspect ratios including 1:1, 3:2, 16:9, 9:16, and 21:9. It also supports 1K and 2K presets via the imageConfig parameter.

How does Nano Banana Pro handle text rendering inside images?

Nano Banana Pro features advanced in-image text rendering with clear, legible multi-language text support—from short captions to long paragraphs. This makes it ideal for posters, infographics, UI mockups, and marketing assets.

Can I edit images conversationally with Gemini 3 Pro Image?

Yes, Nano Banana Pro supports multi-turn conversational editing. Simply ask for changes like 'Make the background a sunset' and the model maintains visual context through Thought Signatures between turns.

What makes Nano Banana Pro different from FLUX 2 Pro or Midjourney?

Nano Banana Pro excels at iterative self-correction, consistent character preservation across multiple edits, and tight Google ecosystem integration. It handles up to 14 reference images for complex multi-image fusion workflows.

Does Nano Banana Pro use Google Search for grounded image generation?

Yes, Nano Banana Pro can use Search grounding to verify facts before generating images. For example, it can fetch current weather data to create an accurate Tokyo weather infographic.

How many reference images can Nano Banana Pro process in one request?

Nano Banana Pro supports up to 14 input images per prompt with a maximum of 7MB per image. It maintains subject and character consistency across up to 5 characters in complex multi-image fusion scenarios.

Nano Banana Pro 的功能

了解 Nano Banana Pro 的核心能力，帮助提升性能与可用性，并改善整体体验。

Nano Banana Pro 的定价

查看 Nano Banana Pro 的竞争性定价，满足不同预算与使用需求，灵活方案确保随需求扩展。

nano-banana-pro（image）

variant / aliasPrice
gemini-3-pro-image (1K/2K)≈ $0.10720
gemini-3-pro-image (4K)≈ $0.19200
gemini-3-pro-image-preview (1K/2K)≈ $0.10720
gemini-3-pro-image-preview (4K)≈ $0.19200
nano-banana-pro-all$0.09600

Nano Banana Pro 的示例代码与 API

获取完整示例代码与 API 资源，简化 Nano Banana Pro 的集成流程，我们提供逐步指导，助你发挥模型潜能。
Python
JavaScript
Curl
from google import genai
from google.genai import types
import os

# Get your CometAPI key from https://api.cometapi.com/console/token, and paste it here
COMETAPI_KEY = os.environ.get("COMETAPI_KEY") or "<YOUR_COMETAPI_KEY>"
BASE_URL = "https://api.cometapi.com"

client = genai.Client(
    http_options={"api_version": "v1beta", "base_url": BASE_URL, "timeout": 600000},
    api_key=COMETAPI_KEY,
)

prompt = "Da Vinci style anatomical sketch of a dissected Monarch butterfly. Detailed drawings of the head, wings, and legs on textured parchment with notes in English."
aspect_ratio = "1:1"  # "1:1","2:3","3:2","3:4","4:3","4:5","5:4","9:16","16:9","21:9"
resolution = "4K"  # "1K", "2K", "4K"

response = client.models.generate_content(
    model="gemini-3-pro-image-preview",
    contents=prompt,
    config=types.GenerateContentConfig(
        response_modalities=["TEXT", "IMAGE"],
        image_config=types.ImageConfig(
            aspect_ratio=aspect_ratio,
            image_size=resolution,
        ),
    ),
)

# Output directory
OUTPUT_DIR = os.path.join(os.path.dirname(__file__), "..", "output")
os.makedirs(OUTPUT_DIR, exist_ok=True)

for part in response.parts:
    if part.text is not None:
        print(part.text)
    elif image := part.as_image():
        output_path = os.path.join(OUTPUT_DIR, "butterfly_4k.png")
        image.save(output_path)
        print(f"Image saved to: {output_path}")

Nano Banana Pro 的版本

Nano Banana Pro 可能存在多个快照，原因包括：更新后保持一致性需要保留旧版、给开发者留出迁移窗口，以及全球/区域端点提供的优化差异。具体差异请参考官方文档。
Model iddescriptionAvailabilityRequest
nano-banana-pro-allThe technology used is unofficial and the generation is unstable etc，Chat format✅Chat format
gemini-3-pro-imageRecommend, Pointing to the latest model✅Gemini generates image
gemini-3-pro-image-previewOfficial Preview✅Gemini generates image

更多模型