生成画像には、アカウントで非表示にできない場合に Veo 3 のサイト表示が入ります。
生成画像のサイト表示を管理できるプランを確認最初の結果がここに表示されます。後のバージョンは「マイ作品」で比較できます。
Create a new still from a prompt, or use references when the subject, product, layout, or place has to remain recognizable.
Nano Banana 2 is built for practical image work: sharper in-image text, better localized layouts, stronger reference handling, and output settings from 1K to 4K. In Veo 3, the generator above gives you the working surface first, then this guide explains when to use the model and how to review the result.
Also visit nano banana 2 for more information.
Start with Nano Banana 2
The model is useful when the result needs more than a pretty frame. It can combine prompt control, reference images, readable layout elements, and higher resolution outputs in one still-image workflow.
Start light for fast creative checks, then raise the resolution for polished stills, product crops, and campaign visuals. The same idea can be reviewed in square, vertical, and wide formats before you commit to a final placement.
Generate a high-resolution still
Nano Banana 2 is a good fit for posters, menu boards, labels, simple infographics, and interface mockups where short text must sit inside the image. Keep the exact words short, name the language, and inspect every character before download.
Create a text-led image
Use a source image when the product shape, material, identity, or approved composition should guide the output. The model can restage the idea while keeping the important reference details visible.
Edit from a reference
The model is designed for prompts that need real-world context, such as maps, locations, factual objects, diagrams, and localized creative. Treat those details as review items: name what must be accurate, then check the result before reusing it.
Plan the image brief
A strong first pass is specific, but not crowded. Give the model one clear job, the required reference or format, and the one detail you will check first.
Keep Nano Banana 2 selected in the generator for prompt-first images. If you already have an approved source, open Image to Image and use the same model for a reference-led edit.
Name the subject, setting, visible text, language, aspect ratio, and output size. For reference work, write what must stay unchanged before describing the new scene.
Check text, identity, composition, and factual details at the intended size. On the next pass, change one instruction so the result improves instead of drifting.
| Capability | Nano Banana 2 | GPT Image 2.5 | Seedream 5.0 |
|---|---|---|---|
| Best-fit image tasks | Grounded stills, readable text, localized graphics, reference edits | Product stills, campaign visuals, local edits, transparent-background work | Design production, region edits, multi-image fusion, dense visual systems |
| Text-to-image | Yes | Yes | Yes |
| Image-to-image editing | Yes | Yes | Yes |
| In-image lettering | Strong fit for short text, signs, labels, and localized layouts | Good fit for posters, labels, ads, and short UI text | Useful for multilingual layouts and graphic-heavy scenes |
| Reference workflow | Multiple references for subject, product, layout, and style direction | Stable reference subjects and precise local changes | Multi-image composition and spatial edit workflows |
| Output planning | 1K to 4K stills with aspect ratio control | 1K to 4K stills with prompt and reference modes | High-resolution design and campaign production |
Use this page when the brief depends on grounded details, readable image text, localization, or a reference-led still.
Use this model →
A strong choice for polished product stills, local edits, reference preservation, and transparent-background image work.
Open the model →Use the previous GPT Image workflow when you want a familiar prompt-to-image baseline or a quick reference edit.
Open the workflow →Start from a written prompt when the subject, composition, and final style are still open.
Open the tool →Upload a source image when a product, person, location, or approved frame should guide the edit.
Open the tool →
Turn a finished still into a short clip once the composition and subject have been approved.
Open the tool →Answers for prompt images, reference edits, text rendering, resolution, and how the model fits into Veo 3.
Nano Banana 2 is an image generation and editing model for still images. It can create from a prompt, work from reference images, and produce higher-resolution outputs for practical design and marketing use. Its strengths are readable image text, localized visual layouts, reference-aware edits, and grounded details that need more context than a purely decorative image.
Use it for campaign stills, posters, simple infographics, product scenes, travel or place visuals, menu-style layouts, UI mockups, and reference-led image variants. The model is especially useful when text and layout matter inside the image. For exact legal copy, prices, dates, or brand-critical typography, generate the visual first and add final text in a design tool.
Yes. Start from Image to Image when you have a source file. Tell the model what must stay fixed, such as the product shape, material, face, layout, or location, then describe the change. Keep the first edit modest. If the subject moves too far, shorten the prompt and put the preserved details first.
Yes. Reference images can guide subject identity, product shape, style, layout, or scene direction. Use fewer references when you need tight control, and add more only when each one has a clear role. A good reference prompt says which image is the source of truth and which details can change.
It is designed to handle short text better than many earlier image models, including localized words, labels, and simple graphic layouts. Keep phrases short, put exact text in quotes, and avoid asking for dense paragraphs. Always inspect the text at the final display size before publishing.
Use 1K for fast idea checks, 2K when the image is close to usable, and 4K for final stills, product crops, or large web placements. Higher resolution is not a substitute for a clearer prompt. If the composition is wrong, fix the prompt before raising the resolution.
Both models can generate and edit still images. Nano Banana 2 is a strong fit when the prompt depends on grounded context, readable text, localization, or multiple references. GPT Image 2.5 is a strong fit for polished product stills, local edits, reference preservation, and transparent-background work. The best model depends on the job, not a single quality score.
The first result appears in the preview area, and completed tasks are stored in My Creations. Use My Creations to compare versions, download approved images, or choose a still for Image to Video. Keeping the prompt, ratio, resolution, and one changed variable with each version makes later edits easier to reproduce.
Veo 3 keeps the model inside the same workspace as your credits, watermark controls, preview, and My Creations library. Create the still, review it, then continue into image editing or image-to-video without rebuilding the brief from scratch.
Generate with Nano Banana 2