Generate and edit Gemini images with prompts you can control.
Gemini Apps can create and edit images in supported countries, languages, accounts, and plans. A useful workflow combines a structured visual brief, small iterative changes, reference-image care, and a final review for factual, visual, copyright, privacy, and disclosure issues.
Build a strong image prompt
Subject and action
Identify the main subject, pose, expression, action, objects, and relationships instead of asking only for a broad theme.
Scene and composition
Describe location, background, camera angle, framing, focal point, depth, negative space, and intended aspect ratio.
Style and light
Specify a medium or visual direction, color palette, lighting, atmosphere, texture, and level of realism without copying a living artist.
Constraints and review
List elements to preserve or avoid, then inspect text, anatomy, identity, logos, diagrams, and factual details before use.
Create an image from text
Open the official Gemini web or mobile app, choose the image experience when the interface provides it, and describe the result as a visual brief. The feature name and underlying image model may change without changing the core workflow.
- Start with ‘create,’ ‘generate,’ or ‘draw,’ followed by the subject and intended use.
- Add composition, setting, style, lighting, colors, mood, and aspect ratio in a readable order.
- If a requirement is essential, state it explicitly rather than relying on the model to infer it.
Edit or combine reference images
Supported Gemini experiences can modify uploaded images or use them as visual context. Explain what must change and what must remain untouched; request one meaningful revision at a time when identity or layout consistency matters.
- Upload only images you have permission to use and avoid exposing private background details.
- Name the exact object, region, color, text, or composition change you want.
- Compare each revision with the original so unwanted changes do not accumulate.
Improve an image through conversation
Treat the first output as a draft. Give concrete feedback about visible differences—such as crop, contrast, expression, spacing, lighting, or missing objects—instead of saying only that the image is wrong.
- Preserve successful elements by listing them before requesting the next change.
- Use a new generation when accumulated edits have damaged structure or consistency.
- Download the full-size result through the official interface and retain the prompt for reproducibility.
Review quality and factual accuracy
Generated images can contain malformed hands, unreadable text, inconsistent reflections, impossible geometry, inaccurate diagrams, misleading maps, or invented product details. Visual polish does not guarantee correctness.
- Zoom in and check anatomy, typography, symbols, logos, shadows, edges, and repeated objects.
- Verify charts, maps, labels, historical details, scientific diagrams, and branded products separately.
- Use conventional design or illustration tools for exact typography and production-critical layout.
Respect rights, safety, and disclosure
Before publishing, consider copyright, trademark, privacy, publicity rights, consent, deception risk, and the rules of the platform where the image will appear. Google’s prohibited-use policy and account terms still apply.
- Do not impersonate a real person or create harmful deceptive media.
- Avoid requesting a direct imitation of a living artist; describe visual characteristics instead.
- Disclose AI generation when the context, audience, employer, client, or platform requires it.