GPT Images 2.5 brings Flare and Sunburst into one workspace for generation and reference edits, with a visible credit quote before each run.
GPT Images 2.5 / Live image workspace
GPT Images 2.5, also written GPT Image 2.5, gives creators a focused image generator with two routes: Flare for everyday generation and one-reference edits, and Sunburst for precision-focused detail work. Write one brief, choose a canvas, review the credit quote, and create in the browser.
Flare and SunburstGeneration and editingOne reference imageVisible credit quote
Two routes, one model family
GPT Images 2.5 is most useful when the workflow is visible. Flare gives you a fast, balanced first direction. Sunburst gives the same brief more room for detail when the composition is worth a precision pass.
Use GPT Images 2.5 Flare when the brief needs a clean first answer without a long setup. It handles product scenes, social layouts, character ideas, and one-reference edits with the same clear workflow.
Move to GPT Images 2.5 Sunburst when the composition is already working and details matter more than speed. It is the route for polished stills, careful edits, and a more finished final frame.
GPT Images 2.5 keeps the prompt, canvas, and reference in the same Voor workspace. Switch between Flare and Sunburst to compare how the same idea resolves without losing your place.
Prompt directions
These images were generated with GPT Images 2.5 Flare on Voor for the prompt directions below. They are original outputs created for this guide, not third-party sample assets. Generate your own image to review the model with your brief.
Use this direction when the image must leave room for a headline, price, or campaign layout.
“Editorial product photograph of a cobalt-blue ceramic coffee cup on warm ivory paper, soft window side light, subtle glaze reflections, generous empty space above for a headline, premium studio still, no text, no logos.”
A vertical canvas shows how GPT Images 2.5 can hold a graphic poster composition with one strong subject and quiet negative space.
“Botanical exhibition poster illustration, one sculptural fern leaf, cream paper texture, forest-green and terracotta palette, abstract geometric frame, clean negative space, high-end print design, no readable text.”
Use an isometric brief when the idea needs a readable scene, tidy geometry, and a warm visual tone.
“Isometric illustration of a quiet creative studio at sunrise, desk, plants, camera, sunlight through tall windows, soft terracotta and sage palette, clean shapes, generous negative space, no text, no logos.”
About these examples: the images on this page were generated with GPT Images 2.5 Flare on Voor. They are our own outputs, not third-party sample files, and they do not carry another platform’s watermark or brand copy. Always review your own result before publishing.
A repeatable prompt structure
GPT Images 2.5 follows a clear brief. Order the prompt like a production note: deliverable, subject, composition, light, material, and the details that must stay fixed on an edit.
Start with the output type, then add one decision at a time. Prompt writing is easier to review when every sentence answers a visible question instead of stacking quality adjectives.
For editing, write the preserved details before the change. That single habit makes GPT Images 2.5 edits more predictable and easier to compare against the source.
Output type + subject
Composition: crop, angle, placement
Light: source, direction, contrast
Material: surface, texture, finish
Exact details: text, objects, colors
For edits: preserve these details firstStart with the output type and the subject. A product photo, editorial poster, or character frame should be obvious in the first clause of the prompt.
Add crop, camera angle, subject placement, and the empty space a headline or logo may need later. Composition language keeps the image useful after generation.
Name the light source, direction, contrast, surface, and texture. GPT Images 2.5 responds well to concrete visual decisions instead of a long list of style names.
For an edit, list the preserved details before the requested change. That order makes it easier to review the result against the source image.
Where it works well
The strongest use cases share the same shape: one clear subject, a visible composition decision, and a result that must be reviewed rather than accepted blindly.
Describe the product angle, material, light, and the empty space needed for campaign copy. GPT Images 2.5 is useful when the composition must survive a layout change.
Use a narrow palette, one strong subject, and generous margins. Add exact copy only when it is short enough to inspect in the final image.
Name the pose, camera distance, wardrobe, and light direction. Keep the same brief while testing one visual variable to build a consistent set.
Upload one supported reference image, preserve identity and geometry first, then describe the change. Review the complete frame before downloading.
| Decision | Flare | Sunburst |
|---|---|---|
| Best for | Everyday generation and fast directions | Detail-focused final passes |
| Canvas | Square, landscape, portrait | Square, landscape, portrait |
| Editing | One-reference edit route | One-reference edit route |
| Output | Medium quality, one PNG | Medium quality, one PNG |
Start with Flare when the idea is still moving. Move to Sunburst when the brief is stable and the image needs a more careful detail pass. Use the live credit quote in the generator to compare the two routes before you run.
Comparing image models? Read the GPT Image 2 page for the earlier generation, try Nano Banana 2 for fast Gemini-style exploration, or start at text to image for the full workspace.
GPT Images 2.5 is an image generation and editing model family used on Voor AI. The workspace exposes Flare for everyday generation and editing, and Sunburst for precision-focused work.
Flare is the balanced starting point for product images, social visuals, and concept exploration. Sunburst spends more time on detail-focused generation and edits when a composition needs a more finished result.
Yes. Choose an edit route, upload one reference image, and describe what should change and what should stay the same. Review the returned image against the source because generative edits can alter more than the requested detail.
GPT Images 2.5 routes on this page support square, landscape, and portrait canvases. The exact canvas is fixed by the selected aspect ratio before generation, and the credit estimate is shown before you run.
The generator displays the exact credit cost for the selected model and mode before submission. Flare and Sunburst use different provider costs, so the quote can change when you switch model or choose an edit route.
Review the current terms and make sure you have rights to any reference image you upload. Inspect generated text, logos, and recognizable people before publishing.
Test Flare for a first direction, then move to Sunburst when the composition is worth a precision pass.
Open the GPT Images 2.5 generatorOptional cookies help us understand usage and measure ads. Essential cookies stay on so Voor works.