匠人学院 JR Academy
Skip to article
Learning guide
Nano Banana Image Editing Guide

Learn image generation and editing workflows with Nano Banana for consistent visual output.

4 chapters·7 guides·Access varies by chapter
Generation vs Editing
On this page

Generation or Editing: Choosing the Workflow

For the Gemini image workflow described as Nano Banana in this guide, choose between creating a new image and changing an existing one before refining prompts.

ModeInputSuitable task
GenerationText, with supported references if availableExploration or a new concept
EditingExisting image plus an instructionPreserve a subject while changing selected details

#Generate when there is no starting asset

Use a short description, inspect the result, add relevant detail and select a direction. This suits concept exploration and clearly identified fictional scenes. Pure generation can be less precise about identity, logos and series consistency.

text
Round 1: A technology-style smartwatch poster in blue. Round 2: Float the watch over a dark-blue background with subtle data lines; keep the product above center and the layout minimal. Round 3: Keep the composition, warm the lower glow and add small light points.

#Edit when preservation matters

Upload the source, identify the exact change, inspect it, refine locally and approve a final version.

text
Keep the product unchanged. Replace the white background with a modern office. Then soften the background to emphasize the product. Finally add a subtle contact shadow below it.

Explicit preservation instructions improve direction but do not guarantee that no other pixels change. Inspect the actual result, especially faces, products and edges.

#Combine the modes

Explore three to five directions, select one and edit it. Alternatively, prepare a core subject before creating its environment, or use a reference to establish the visual direction of new content.

#A decision sequence

With an existing asset and a small change, begin with editing. For a substantially different result, use generation with the asset as a clearly defined reference if supported. Without an asset, use a detailed brief for an established target or a short exploratory brief when the direction is still open.

#Example: social cover

Generation can create a blue/purple AI concept cover with a central abstract network and left title space. Editing can add the network to an already approved background while preserving the empty left third. The first offers exploration; the second offers a controlled starting background.

#Troubleshooting

If editing changes protected areas, specify the exact element and position, and use supported annotations/masks where available. If generation varies, fix palette/style/composition and use an approved reference. Save outputs so either mode can become the input to the next stage.

#Next steps

Published by
JR Academy
Learning resources

Individual authors and reviewers are credited where recorded.

Free Resources

Curated free tools, courses, and study materials

Find practical learning resources in one place.

Browse Free Resources →

Related Roadmaps

FAQ

Nano Banana 是什么?
Nano Banana 是 Google 基于 Gemini 模型推出的 AI 图像编辑工具,支持对话式编辑,用自然语言描述即可完成复杂的图像操作。
Nano Banana 和 Midjourney 有什么区别?
Nano Banana 更擅长对话式编辑和改图,Midjourney 更擅长从零生成。Nano Banana 对中文支持更好,且可在 Gemini App 中免费使用基础功能。
Nano Banana 有哪些模型可选?
三个层级:Nano Banana Fast(快速预览)、Nano Banana Pro(高质量输出)、Nano Banana Ultra(专业级 4K)。
如何提高风格一致性?
使用多图参考、建立风格关键词清单、每次都引用第一张图作为风格锚点,并减少单次描述的变化量。

Done reading? Next steps