Structure a prompt
Draft up to 1,200 characters and use the live counter to keep the request focused. Prompt text stays in the current browser page.
Browser-only prompt preview
Nano Banana image generator prompt workflow
Draft a structured image prompt and inspect the planned text-to-image and permitted-reference workflow. This preview does not generate, upload, or download images while provider and safety controls remain unapproved.
Try Nano Banana is an independent third-party service and is not affiliated with or endorsed by Google.
Provider review in progressYour result will appear here
The result panel and download action will activate only after the server-side provider, content lifecycle, and failure handling pass review.
ValidationModerationTimeoutRate limitProvider errorCleanup
Current state: prompt drafting happens only in this browser. Nothing is sent to a model, and the page does not accept a reference file.
Available now
What you can do in the Nano Banana image generator preview
Use the page as a private drafting aid while the server-side generation workflow remains gated.
Compare input paths
Learn when a text-only request is sufficient and when a permitted reference could add necessary shape or composition information.
Plan the review
Check the planned validation, moderation, timeout, provider-error, cleanup, result-review, and download states before trusting a future workflow.
Not available: model submission, reference upload, image generation, result storage, and download. No provider, model, quota, speed, resolution, retention period, or commercial-use promise has been approved.
How the workflow is designed
One request, one clear terminal state
A generation should never leave you guessing whether it finished, failed, or is still using resources.
The planned Nano Banana image generator accepts a required prompt and, when permitted, one optional reference image. Before provider work begins, the server must validate the request, apply rate and abuse checks, and make a moderation decision.
An accepted request must record exactly one final outcome: success or a controlled failure. The interface then shows a validated result and download action, or a useful error state that does not expose private prompt content, provider payloads, or raw system errors.
No provider or model is selected. This page makes no claims about speed, resolution, free quotas, data retention, commercial rights, training use, watermarking, or regional availability.
Prompt anatomy
Write instructions that describe visible choices
Start with the result you need, then add only the details that affect composition or review.
Subject and action
Name the main subject and what is happening. “A ceramic cup beside an open notebook” is easier to inspect than “a creative productivity scene.”
Frame and light
State the viewing angle, crop, depth, light direction, and amount of empty space. These choices determine where the viewer looks first.
Materials and constraints
Add relevant surfaces, colors, required objects, and exclusions. Do not fill the prompt with synonyms that do not change the picture.
Representative prompt, not a generated result
“Create a 4:5 editorial product photograph of a plain amber glass bottle on a pale stone block. Use soft side light from the left, a warm off-white background, a low eye-level camera, and open space above the bottle. Keep the label area blank. No logo, readable text, hands, extra containers, or watermark.”
This brief identifies format, subject, material, position, light, background, camera, text-safe space, and exclusions. Each part can be reviewed separately.
Text-to-image or reference editing
Choose the smallest input that can do the job
Start from text
Use prompt-only generation when the subject and composition can be described without an existing image. This reduces the amount of source material shared and avoids accidental dependence on a protected or personal reference.
- Describe the intended use and crop.
- Put the main subject before minor styling details.
- Reserve space for text instead of asking for final copy.
- Review visual facts before publishing.
Edit a permitted reference
Use a reference only when it carries information that text cannot express efficiently, such as an approved product shape or owned composition. Confirm rights and remove unrelated personal data first.
- State what must stay unchanged.
- Name the exact change you want.
- Avoid real-person content until applicable rules allow it.
- Check the whole image, not only the edited area.
Failure and recovery
Useful feedback at every stop
Before a request
Empty prompt, invalid file, oversized file, unsupported type, rate limit, and moderation feedback should appear before paid provider work starts.
During processing
The interface needs clear loading, bounded waiting, timeout, cancellation, and provider availability states. A retry must never duplicate a hidden accepted job.
After processing
A success needs a validated result and download. A failure needs a controlled reason, safe retry guidance, and cleanup without leaking private input.
Image review checklist
Inspect the result before download or publication
- Prompt fit: does the image show the requested subject, action, setting, and crop?
- Structural quality: inspect hands, edges, reflections, repeated objects, shadows, and perspective.
- Text and marks: check every label, sign, symbol, signature-like mark, and watermark.
- Rights and safety: look for unwanted likeness, private details, trademarks, or prohibited content.
- Final context: test the actual placement, responsive crop, contrast, and alternative text.
For a more detailed method, read how to generate AI images from a creative brief. Marketing teams can use the campaign visual brief.
Questions about the tool
What the preview does and does not do
Can I generate an image on this page today?
No. The page currently helps you draft a prompt and understand the planned workflow, but it does not send requests to an image model. Generation will remain disabled until provider, moderation, privacy, upload, retention, rate-control, and recovery requirements are approved and implemented.
Why is the generate button disabled?
The product has not approved a provider, model, moderation policy, upload limits, retention rules, or recovery behavior. Disabling submission prevents the preview from accepting content without those safeguards.
Does the prompt leave my browser?
No. In this preview, the text area only updates its local character count and a content-free interaction event. Prompt text is not submitted, stored, logged, or included in analytics.
Which image model will the tool use?
No model is selected. Candidates must pass a fixed benchmark and a current review of provider identifiers, terms, safety, data use, retention, price, quotas, regions, and deprecation policy.
Will generated images be safe to publish?
No system can replace human review. A user must still check factual accuracy, visual defects, privacy, likeness, intellectual-property risk, platform rules, and suitability for the intended audience.
Learn the complete process in the AI image generation guide, review naming and provider claims in the Nano Banana model guide, or use the evidence-based generator comparison checklist.