Nano Banana 2.1 is here | Pinpoint edits, consistent characters.

GPT Image 2.6 Sunburst Prompt Guide: Better Text, Product Photos and Ad Creatives

GPT Image 2.6 Sunburst Prompt Guide: Better Text, Product Photos and Ad Creatives
Learn how to write GPT Image 2.6 Sunburst prompts for readable text, accurate product photos, reference-image edits and multi-format ad creatives.
Create with GPT Image 2.6 Sunburst

A useful GPT Image 2.6 prompt is a production brief, not a pile of style words. State the asset you need, the subject that must remain accurate, the text that must be readable, the composition and the delivery format. This guide gives you a repeatable workflow for text-heavy images, product photos, reference-image edits and ad variants.

Quick answer

For reliable results, write prompts in five parts: task, subject, locked details, composition and output. If you upload a reference image, list what may change and what must not change. Generate a small first batch, inspect it at the final display size, then revise one defect at a time instead of rewriting the whole brief.

What GPT Image 2.6 Sunburst is useful for

The current iMini creation page accepts either a written prompt or a reference image. It provides PNG output, a high-quality option and automatic background handling. That makes it suitable for generating a new visual from a brief or editing an existing asset while keeping the original as the source of truth.

TaskGive the modelLock firstCheck before use
Text-led imageExact approved copy and hierarchySpelling, punctuation, line breaksRead every character at final size
Product photoOriginal packshot and scene briefShape, label, logo, color, materialCompare against the original product
Ad creativeOffer, audience, channel and ratioProduct and approved claimsSafe area, crop and CTA space
Reference editOriginal image plus one requested changeIdentity, pose, camera angle or packagingInspect changed and unchanged regions

Use a five-part prompt formula

Start with the deliverable, then narrow the model's freedom. A practical formula is: Create or edit [asset] for [use]. Show [subject]. Preserve [locked details]. Change [allowed details]. Compose it as [layout, lighting and ratio]. Output [format and quality]. Put approved words in quotation marks so they are easier to compare with the result.

Create a [type of image] for [channel or use].
Subject: [person, product or scene].
Must preserve: [identity, logo, label, color, material, camera angle].
Change only: [one permitted change].
Composition: [framing, background, lighting, focal point, negative space].
Exact text: "[approved copy]". Do not add other text.
Output: [aspect ratio], clean edges, high detail, PNG.

Make text readable without sacrificing the layout

Separate the copy from the visual direction. Specify the exact headline, price or short label, then describe its hierarchy and placement. Avoid asking the model to invent campaign copy while it is also composing the image. For long legal text, changing prices or multilingual campaigns, create a clean background and finish the copy in an editable text layer.

Design a clean promotional image for [product].
Use this exact headline: "[HEADLINE]".
Use this exact subheading: "[SUBHEADING]".
Place the headline in the upper-left safe area with strong contrast and generous spacing.
Keep the product, logo, label and packaging unchanged.
Do not add, paraphrase or misspell any text.
Leave a clear area for a separate call-to-action button.

Build accurate product photos

For product work, accuracy outranks decoration. Upload the cleanest packshot available and name every feature that carries commercial meaning: package silhouette, logo, label copy, cap, material, finish, color and proportions. Describe the new environment only after those preservation rules.

A good workflow is to make one neutral hero image first, approve the product, and then branch into seasonal or channel-specific scenes. This prevents a visual mistake from being copied across an entire campaign.

Edit the uploaded product photo into a premium studio hero image.
Keep the product shape, proportions, logo, label text, packaging color and material exactly unchanged.
Replace only the background with [scene].
Use [lighting direction] with a realistic contact shadow and natural reflections.
Center the product with [amount] negative space for copy.
Do not invent accessories, claims, badges or extra packaging.
Output a high-detail PNG in [ratio].

Create ad variants without redesigning the product

Do not request every placement in one crowded instruction. Approve a master composition, then make one ratio at a time. Tell the model which element is the anchor, where empty space belongs and which edges may be extended. For paid ads, keep offer wording and compliance claims outside the model's improvisation.

Using the approved reference image, create a [1:1 / 4:5 / 9:16] ad variant for [platform].
Keep the product, label, logo, people, colors and lighting consistent with the reference.
Recompose the scene for the new ratio without stretching or cropping the product.
Reserve a clean safe area at [location] for the headline and CTA.
Do not add text, badges, prices or new objects.
Maintain realistic scale, shadows and perspective.

Reference-image editing workflow

Keep the original file outside the editing chain. For each revision, return to the approved source or the last verified version. Ask for one meaningful change, compare at full size, and save the result with a version name that records what changed. If identity or a product label drifts, stop and repair that region before making another creative change.

Review pointPass conditionIf it fails
Subject accuracyFace or product matches the sourceReturn to the clean reference and reduce the edit area
TextEvery character matches approved copyEdit one text block or use an editable layer
LightingShadow direction and reflections agreeSpecify one light direction and contact shadow
CompositionImportant elements survive the final cropName the safe area and generate each ratio separately
ExportCorrect dimensions, PNG and clean edgesRegenerate or resize from the approved master

Common prompt mistakes

Avoid vague requests such as “make it better,” conflicting style references, several unrelated edits in one turn and unsupported promises such as guaranteed perfect text. Do not approve a thumbnail without checking the full-size file. For commerce, also compare the image with the actual item so attractive output does not become misleading product information.

Related guides

Use the GPT Image 2.6 vs 2.5 vs 2 comparison when you are choosing a version. For an older workflow and terminology, see the GPT Image 2.5 guide. If preserving packaging is the main problem, follow the product-photo editing checklist.

FAQ

How do I write a good GPT Image 2.6 prompt?

Name the asset, subject, locked details, allowed change, composition and output. Put exact text in quotation marks and make one major change at a time.

Can I use a reference image?

Yes. The current iMini page accepts a reference image. Keep the original and specify what must remain unchanged before describing the edit.

How can I keep a product label unchanged?

Use the original packshot, list the label and packaging as locked details, restrict the edit to the background, and compare the result with the source at full size.

Should I generate all ad sizes at once?

Approve one master first, then generate each ratio separately. This gives you better control over crops, safe areas and product scale.

What if the image text is still wrong?

Edit one short text block at a time. If exact copy remains unreliable, generate a text-free image and add the final words in an editable design layer.

Turn your prompt into a finished image

Models used in this guideGPT Image 2.6 SunburstExplore modelStart creating