- 最高 4K 解析度輸出（512px / 1K / 2K / 4K 四階）
- 10+ 種長寬比，包括 21:9、1:4、8:1 等
- 精準清晰的圖中文字渲染
- 接近 Pro 級畫質（約 95%），Flash 級速度
- 跨場景最多 5 個角色的一致性維持
- 單一工作流程中最多 14 個物件的保真度
- 透過自然語言進行精準編輯（移除、替換、調整姿態）
- 多圖融合與無縫合成
Nano Banana 2 新功能
比 Pro 快 3-5 倍採用 Gemini 3.1 Flash 架構，Nano Banana 2 標準影像生成僅需 4-8 秒，而 Nano Banana Pro 需要 10-20 秒。
影像搜尋增強NB2 的亮點功能 — 生成過程中可透過 Google 搜尋擷取真實世界參考影像，大幅提升地標、名人和品牌標誌的生成準確度。
精準文字渲染為行銷素材、賀卡和在地化內容生成準確清晰的文字，甚至可以在影像中直接翻譯和在地化文字。
多角色一致性跨場景維持最多 5 個角色和 14 個物件的視覺一致性 — 非常適合分鏡、漫畫創作和行銷活動。
Text Rendering
A minimalist coffee shop promotional poster with the text 'MORNING BREW — Fresh Roasted Daily' in elegant serif font, warm earth tones, steam rising from a ceramic cup, clean layout with plenty of whitespace
Character Consistency
A young woman with short red hair and freckles, wearing a green jacket, standing in a rainy Tokyo street at night with neon reflections on wet pavement, cinematic lighting, photorealistic
Photo to Action Figure
Transform the person in the photo into an action figure, styled after [CHARACTER_NAME] from [SOURCE / CONTEXT]. Next to the figure, display the accessories including [ITEM_1], [ITEM_2], and [ITEM_3]. On the top of the toy box, write "[BOX_LABEL_TOP]", and underneath it, "[BOX_LABEL_BOTTOM]". Place the box in a [BACKGROUND_SETTING] environment.
Search Grounding
A photorealistic aerial view of the Eiffel Tower at golden hour, with the Seine River winding through Paris below, warm sunset light casting long shadows, high detail, 4K resolution
Product Photography
A frosted glass perfume bottle with a marble cap on a white marble surface, soft studio lighting from the left, subtle reflections, minimalist luxury aesthetic, product photography style
Style Transfer
Transform this photo into Studio Ghibli animation style, keeping the same composition and subjects, lush watercolor backgrounds, soft diffused lighting, whimsical atmosphere
4K Output
A cozy Japanese ramen shop interior at night, steam rising from bowls, warm amber lighting, detailed wooden counter with various condiments, a chef working in the background, 4K, ultra detailed
為什麼選擇 Nano Banana 2？
⚡
Flash 級速度
比 Nano Banana Pro 快 3-5 倍，標準生成時間僅 4-8 秒🎯
接近 Pro 級畫質
在大多數情境下可達到 Pro 約 95% 的畫質水準💰
更具性價比
成本約為 Nano Banana Pro 的一半 — 讓高品質 AI 影像生成更加普及技術規格
架構：Gemini 3.1 Flash (GEMPIX2)
解析度支援：512px 至 4K（512px / 1K / 2K / 4K 四階）
長寬比：1:1, 4:3, 3:4, 2:3, 3:2, 16:9, 9:16, 1:4, 4:1, 8:1, 21:9
一致性：單一工作流程最多 5 個角色 + 14 個物件
內容安全：SynthID 浮水印，相容 C2PA 標準
API 存取：Gemini API、Vertex AI、AI Studio、Gemini CLI
立即體驗 Nano Banana 2
Pro 級畫質 Flash 級速度 — 輕鬆建立具有角色一致性、文字渲染和 4K 解析度支援的精美視覺內容。
✨免費額度立即開始
⚡即時 API 存取
🌐無需任何設定
Google Nano Banana 2 Reference to Image
Nano Banana 2 Reference to Image (Gemini 3.1 Flash Image) is Google's advanced AI-powered video-to-image generation model, designed to generate high-quality static images from video clips combined with text instructions. Built on the same cutting-edge model as Nano Banana 2 Edit, it adds the ability to use video content as a rich reference source — extracting visual context, themes, and key frames to synthesize new images with precision and semantic awareness.
This model is ideal for creating thumbnails, posters, promotional artwork, and scene summaries by leveraging the visual richness of existing video content alongside natural language guidance.
Why Choose This?
-
Video as reference — Provide a video clip (HTTP URL or YouTube URL) and let the model extract its visual context to guide image generation.
-
Multi-image reference — Optionally upload up to 10 additional reference images to complement the video input for complex compositions.
-
Natural language control — Describe exactly what you want with a text prompt; the model understands context, themes, and relationships from both the video and text.
-
Thinking levels — Choose how much internal reasoning the model applies — higher thinking levels improve quality on complex tasks.
-
Media resolution control — Balance detail and token usage for input video frames with LOW, MEDIUM, or HIGH media resolution modes.
-
Web & image search grounding — Optionally enable real-time web or image search to enrich generation with current information.
-
Multi-resolution output — Generate at 1K, 2K, or 4K resolution.
-
Flexible aspect ratios — Multiple options including 1:1, 3:2, 2:3, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, and 21:9.
-
Format choice — Export in PNG or JPEG format.
How It Works
The model analyzes your video clip by sampling frames at the specified FPS rate, then interprets the visual content within its multimodal context window. Combined with your text prompt and any additional reference images, it synthesizes a new image grounded in the video's themes, style, and key visual elements. This makes it especially powerful for creating content that is visually consistent with existing video assets.
Parameters
| Parameter | Required | Description |
|---|
| prompt | Yes | Text description of the desired output image |
| video_clips | Yes | Source video clip(s) for reference generation (max: 1, see below) |
| images | No | Additional reference images (max: 10, click "+ Add Item" to add more) |
Video Clip Fields
| Field | Required | Description |
|---|
| url | Yes | URL of the source video clip. Supports HTTP URL or YouTube video URL. HTTP video is limited to 15MB. |
| start | Yes | Start time in seconds for trimming the video clip (min: 0) |
| ends | Yes | End time in seconds for trimming the video clip. Set 0 to use the whole video. |
| fps | Yes | Frame sampling rate (FPS) of the video clip. Range: 0–24. Lower values reduce token usage. |
Output Options
| Parameter | Required | Description |
|---|
| aspect_ratio | No | Aspect ratio: 1:1, 3:2, 2:3, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 21:9 |
| resolution | No | Output resolution: 1k (default), 2k, 4k |
| output_format | No | Output format: png (default), jpeg |
Advanced Options
| Parameter | Required | Description |
|---|
| thinking_level | No | Reasoning depth: default, high, minimal. Higher levels improve quality on complex tasks but increase latency. |
| media_resolution | No | How input media frames are processed: default, low, medium, high. LOW reduces tokens per frame, allowing longer videos. |
| enable_web_search | No | If enabled, grounds generation with real-time web information. |
| enable_image_search | No | If enabled, grounds generation with real-time image search results. |
How to Use
- Provide a video clip — enter the video URL (HTTP or YouTube) and set start/end times and FPS sampling rate.
- Write your prompt — describe the output image clearly (e.g., "Create a cinematic poster based on the key scenes in this video").
- Add reference images (optional) — upload additional images to guide composition or style.
- Choose aspect ratio (optional) — select a preset or leave empty for default.
- Select resolution — choose 1K, 2K, or 4K based on your quality needs.
- Choose output format — PNG for transparency support, JPEG for smaller file size.
- Adjust advanced settings (optional) — set thinking level, media resolution, or enable search grounding.
- Run — submit and download your generated image.
Pricing
The total cost is determined by the output image resolution multiplied by the number of output images, plus optional per-request fees for video clip input, web search, and image search grounding.
SKU Prices
| SKU | Description | Unit Price |
|---|
| sku_1k | 1K resolution output image | $0.08 |
| sku_2k | 2K resolution output image | $0.12 |
| sku_4k | 4K resolution output image | $0.16 |
| sku_video_clip | Video clip input (per request) | $0.07 |
| sku_web_search | Web search grounding (per request) | $0.014 |
| sku_image_search | Image search grounding (per request) | $0.014 |
cost = (resolution == "2k" ? sku_2k : (resolution == "4k" ? sku_4k : sku_1k)) * images
+ (enable_web_search ? sku_web_search : 0)
+ (enable_image_search ? sku_image_search : 0)
+ (len(video_clips) > 0 ? sku_video_clip : 0)
Examples:
| Resolution | Video Clip | Web Search | Image Search | Total Cost |
|---|
| 1K | Yes | No | No | 0.08+0.07 = $0.15 |
| 2K | Yes | No | No | 0.12+0.07 = $0.19 |
| 4K | Yes | No | No | 0.16+0.07 = $0.23 |
| 1K | Yes | Yes | No | 0.08+0.07 + 0.014=∗∗0.164** |
| 1K | Yes | Yes | Yes | 0.08+0.07 + 0.014+0.014 = $0.178 |
| 1K | No | No | No | $0.08 |
| 2K | No | No | No | $0.12 |
| 4K | No | No | No | $0.16 |
The video clip fee (0.07),websearchfee(0.014), and image search fee ($0.014) are each charged once per request when the respective feature is enabled, regardless of content volume.
Best Use Cases
- Video Thumbnail Generation — Automatically create compelling thumbnails that reflect the video's content and mood.
- Promotional Posters — Generate movie-style or campaign posters grounded in actual video footage.
- Scene Summarization Art — Create visual summaries or highlight artwork from long-form video content.
- Brand Content Creation — Produce consistent image assets from brand video campaigns.
- Educational Infographics — Transform instructional videos into static visual materials.
- Social Media Assets — Generate platform-optimized images (vertical, square, landscape) from video content.
Pro Tips
- Use low FPS (0.5–1) for long videos to keep token usage within limits while still capturing key frames.
- Set precise start/end times to focus the model on the most relevant segment of your video.
- Combine specific text prompts with the video input — vague prompts may produce generic results.
- Add reference images alongside the video to guide composition style more precisely.
- Use
thinking_level: high for complex scene interpretations or when visual fidelity matters most.
- Set
media_resolution: low when analyzing long videos to allow more frames within the context window.
- 2K offers excellent quality at a reasonable price — only $0.04 more than 1K per image.
- YouTube URLs are supported directly — no need to download and re-upload public videos.
Notes
- Both
prompt and video_clips are required fields.
- Maximum video clips: 1 per request.
- HTTP video URLs are limited to 15MB; use YouTube URLs for larger videos.
- Maximum additional reference images: 10.
- FPS range: 0–24. Higher FPS captures more frames but consumes more tokens.
- The video clip fee ($0.07) is a flat per-request charge, not per frame or per second.
- If
aspect_ratio is not selected, the model uses a default ratio.
- 4K resolution costs 2× the standard 1K rate.
- Ensure your content and prompts comply with Google's Safety Guidelines.
