OPENAI · A NEW CHAPTER IN IMAGE CREATION
Imagine freely. Refine precisely.
Turn a product photo into a campaign, a rough sketch into a scene, or a portrait into a new look. GPT Image 2.5 brings generation and reference editing into one workspace. Describe what you want to change, keep the details that matter, and build on your favorite result.
Start free without signing in · Sign in to keep creating
Start free without signing in · Sign in to keep creating
SEE THE POSSIBILITIES
Compare the references and results, then inspect the prompt actually published with each image. These six examples cover sketches, local edits, composition, lighting, portraits and nature. Select an image to open it in full.
Sketch to architecture
The red doodle is the creator’s actual reference; the building is the reported output. Compare its silhouette with the curved glass volumes and elevated terraces. This pair shows sketch-led architecture, rather than an unrelated finished building.
See this image’s prompt
Interior editing
Compare the pale sofa with the forest-green velvet edit, checking the furniture, framing and light. This is the first pass in the editing sequence below. The published instruction asks for new upholstery, without adding decorations.
See this image’s prompt
Composition from a sketch
This edit uses two references: the original supplies the figure, purple suit and red seat; the white-line sketch supplies the reclining pose and composition. The result sits below both. Upload both references when adapting this prompt.
See this image’s prompt
Objects and lighting
The creator’s full comparison runs from the original at the top to GPT Image 2.5 in the middle and GPT Image 2.0 below. Look at the tabletop shadows, metal reflections and mirror after the lamp moves right. Original labels are retained; this is a creator example, not a Nano Banana 2.1 benchmark.
See this image’s prompt
Everyday portraits
The image accompanies a short request for a realistic iPhone photo of a woman in a cafe. The coffee, counter and ordinary indoor lighting match that brief. This is a text-to-image example, without a supplied portrait or an identity-preservation comparison.
See this image’s prompt
Natural scenes
The published prompt asks for a clearing, plentiful green foliage and detail. The result centers on the canopy, undergrowth and light entering the woods. Enlarge it to inspect textures and noise for yourself.
See this image’s promptONE SCENE. FOUR DECISIONS.
The useful question is often “can I change this without losing that?” With GPT Image 2.5, separate your brief into the change you want and the details you want to preserve. Start with the room, change the upholstery, add artwork, then adjust the mood. Each revision becomes a clear decision you can compare with the version before it.
A neutral sofa in a sunlit room.
To try this with your own image, upload it above, name one change, and use “Continue editing” on the result. Keep the best version as your next reference. Compare window geometry, furniture placement and earlier edits after each step; preservation is a goal to check, not a pixel-perfect guarantee.
Choose around the work in front of you. Both GPT Image 2.5 versions accept text and reference images, and both can continue an edit. Flare is the starting point for exploring a direction; Sunburst is an option when a demanding composition deserves a closer pass. Try the same brief and references when comparing them.
FOR EXPLORING IDEAS
Start here for social posts, everyday product images, mood exploration and early drafts. Make one useful variation, inspect it, and decide what to change next. A clear brief at standard quality is a practical way to test your composition before spending more credits on a larger or higher-quality output. Generation time varies with your settings and the current queue.
FOR THE FINISHING TOUCHES
Try this for a campaign hero, a detailed product arrangement or a revision where several visual constraints need to hold together. Describe the exact material, perspective and preserved elements you care about. Allow more time for the result, then compare it with your reference at full size. Extra precision is useful when it improves a detail that matters to your final image.
Dimensions depend on aspect ratio, up to 8.3 megapixels. The 16:9 and 9:16 formats support a 3840-pixel long edge.
Each image is paired with its published English prompt. Copy keeps the original wording. Editing examples also need the corresponding reference images; your own inputs will lead to different results.
Original prompt · Editing example: upload the corresponding reference images shown above.
turn it into a future building
Original prompt · Editing example: upload the corresponding reference images shown above.
Reupholster the sofa in a deep forest green velvet. Change nothing else in the room.
Original prompt · Editing example: upload the corresponding reference images shown above.
change the composition to this
Original prompt · Editing example: upload the corresponding reference images shown above.
Move the lamp to the right
Original prompt · Text to image; no reference required.
make me a realistic iphone photo of a woman in a cafe
Original prompt · Text to image; no reference required.
Photo of a clearing in the woods with lots of green foliage, highly detailed
Create product images, posters, portraits and illustrations, or edit photos you already have.
No. Describe the change in everyday language and say what should stay the same.
Yes. Start free without signing in, then continue with available free credits. Free generations use Flare at 1K Standard with one reference image.
Costs vary by model, resolution and quality. With paid access, you can see the exact cost before generating.
Start with Flare for the free experience. Sunburst requires paid access.
Guest and free-credit generations support 1 reference image. Paid access supports up to 5 where the selected model allows it.
Yes. Supported 2K or 4K settings require paid access and general credits.
Yes. You can download the guest result. Sign in after the successful first image to link and save it to your account.
Bring a photo, a sketch or a few clear sentences to GPT Image 2.5. Choose a direction, create a first version and refine the details that make it yours. Your next campaign, portrait or visual story can start right here.
Create with GPT Image 2.5
