GPT Image | v2.6 | Sunburst | Edit.

GPT Image | v2.6 | Sunburst | Edit Overview

GPT Image | v2.6 | Sunburst | Edit provides OpenAI image editing through each::labs. Built on GPT Image 2.6 Sunburst, it is intended for detailed compositions and precise visual revisions. Supply one or more source images and describe the changes you want. The model interprets your instructions together with those references to produce a new image, making it useful for background changes, visual restyling, and combining subjects into a coherent scene. You control output dimensions, rendering quality, background behavior, and file format using explicit request parameters. The endpoint supports PNG, JPEG, and WebP rather than every format available in the broader provider family. Start with one image and automatic quality, review the result, and refine your brief before generating more variants.

Capabilities

• Transform existing images using a natural-language editing brief.
• Combine instructions with up to 16 source image URLs.
• Specify scene layout, lighting, materials, and short visible text in the prompt.
• Select square, portrait, or landscape output presets.
• Adjust rendering quality separately from planning complexity.
• Request transparent backgrounds with PNG or WebP output.
• Produce up to ten image outputs per request for reviewing alternatives.

Use Cases for GPT Image | v2.6 | Sunburst | Edit

Marketers can adapt a product photograph to a seasonal campaign: “Keep the bottle shape and label; add a warm autumn tabletop setting.” This uses reference-guided editing to produce a new presentation without starting from a blank prompt. Designers can explore material treatments: “Preserve the chair silhouette; change the upholstery to dark green velvet.” They should compare the result against the source before treating it as a product representation. Creators can reframe editorial imagery: “Keep the subject recognizable; replace the background with a softly lit studio wall.” Explicit preservation instructions make the intended revision clearer. Developers can build an image-revision interface around source URLs, a change brief, and output-format controls. Present returned candidates for review, retain the original files, and let people submit a narrower follow-up edit when details drift.

Tips and Tricks

For GPT Image | v2.6 | Sunburst | Edit, structure the prompt around subject, setting, composition, lighting, and the exact change or result you want. Refer to source images by their position and state preservation requirements before new instructions. Avoid asking for several incompatible changes to the same feature. For example: “Keep the product and label unchanged; replace only the background with pale blue paper and soft studio shadows.” “Use image 1 for the subject and image 2 for the object; place the object on the subject’s open palm in natural afternoon light.” Compare one change at a time. Use low or medium complexity for focused experiments, and test higher complexity for dense layouts. Choose an explicit size preset when dimensions matter, then inspect the delivered image rather than assuming every requested detail is exact.

Technical Specifications

• Inputs: a required text prompt, up to 32,000 characters, plus 1–16 source image URLs.
• Dimensions: image_size defaults to 1024x1024. Listed presets include 1536x1024, 1024x1536, 2048x2048, 2048x1152, 3840x2160, and 2160x3840; auto is also available.
• Quality: low, medium, high, xhigh, max, or auto. The default is auto.
• Batch size: 1–10 output images; the default is one.
• Formats: PNG by default, with JPEG and WebP options; transparent backgrounds require PNG or WebP.
• Planning: complexity accepts low, medium, or high; omitted complexity uses high. No fixed processing-time guarantee is published for this endpoint.

Things to Be Aware Of

Generation is not deterministic: repeated requests can produce different compositions and details. Faces, logos, object geometry, and small text can change even when the prompt asks to preserve them. Reference URLs must remain accessible during processing; a broken or unavailable image can prevent a successful edit. Automatic quality is not a fixed-price setting. Larger images and more outputs can increase usage. The catalog marks resolutions above 2560x1440 as experimental. Standard content filtering remains active, including when the compatibility moderation value is low.

Key Considerations

Choose GPT Image 2.6 Sunburst when your workflow calls for detailed compositions and precise visual revisions. Use clear source photographs and identify which image supplies each subject or object. A descriptive prompt should distinguish elements that must remain from elements that may change. Billing is usage-based, not a fixed price for every image. Dimensions, rendering quality, and image count affect the final charge. Start with one output, keep quality automatic while exploring, and compare a small number of candidates before committing to a larger batch.

Limitations

This endpoint returns raster images, not SVG files, animated GIFs, video, or editable design layers. It does not expose a seed or a separate negative-prompt field. It does not expose a mask parameter; request targeted edits in the prompt, without expecting pixel-exact preservation. Exact text, identity, and geometry are not guaranteed. Output quality depends on the brief. Select an explicit listed size rather than custom when predictable dimensions are required.