Create from a detailed brief
Describe composition, lighting, materials and any exact text. Use Flare to explore campaign images, editorial scenes and design concepts.

Describe your idea below to create your own image.
0/4000
Up to 10 MB each. References are resized to a maximum 1024px edge. Each reference adds credits.
4K uses up to 8.29 megapixels; square output is 2880 × 2880.

Start from a written brief or build on your own images. Choose a model and quality level for each task, then give clear instructions about the details that matter.
Describe composition, lighting, materials and any exact text. Use Flare to explore campaign images, editorial scenes and design concepts.
Use references to guide your subject and composition. Explain what should stay and what should change; try Sunburst for edits that need closer attention.
Add a transparent PNG mask matching your first reference. Transparent areas identify the region to regenerate, while your prompt explains the requested change.
Choose Low through Max separately from resolution. Higher quality costs more; the displayed quote includes your prompt and uploaded references.

Official EvoLink Playground sample for GPT Image 2.5 Flare.

Official EvoLink Playground sample for GPT Image 2.5 Sunburst.

A 1K Medium Flare text-to-image result generated and saved by this workspace.

A 1K Medium Sunburst edit from a normalized reference image, saved by this workspace.
GPT Image 2.5 is OpenAI’s image generation and editing family, with two models: Flare for faster everyday creation and Sunburst for detailed edits where precision matters more than speed.
This workspace supports text prompts, up to 16 reference images, an optional PNG editing mask, 15 aspect ratios and five quality levels. Create one image per task, then download PNG, JPEG or WebP. References are resized to a maximum 1024px edge before generation.
Give the model a deliverable, exact content and visual priorities. These starting points make it easier to review the result and decide what to change next.
Specify the headline, product, placement and visual hierarchy. Keep essential wording concise and proofread it in the generated image.
Upload a product reference and request a new setting, lighting or layout. Name the shape, colors and other details that should remain recognizable.
Describe panels, labels, audience and information order. Use the result as a visual concept and verify text, numbers and diagram relationships.
Create, review and download on one page. Your task stays in recent history so you can return while generation is running.
Describe the subject, layout, exact words and visual style. For edits, separately list the elements to preserve and change.
Choose Flare or Sunburst, optionally add references and a mask, then select resolution, ratio and format. Review the total credit cost.
Start the task, review the completed image and download it. Return to recent tasks on this page to continue checking a running generation.
The two variants offer the same workspace controls. Start with Flare for everyday generation; compare Sunburst when edit fidelity is your priority.
| Compare | GPT Image 2.5 Flare | GPT Image 2.5 Sunburst |
|---|---|---|
| Primary focus | Faster everyday generation and editing. | More precision for detailed creative edits. |
| Generation time | The faster variant in OpenAI’s positioning. | Can take longer for precision-focused work. |
| Workspace controls | 1K/2K/4K, five quality levels, 15 ratios. | The same resolution, quality and ratio controls. |
| Image editing | Up to 16 references and an optional PNG mask. | Up to 16 references and an optional PNG mask. |
| Site credits | Calculated from quality, resolution, prompt and references. | The same site quote for the same settings. |
Answers about the models, reference editing, resolution and credit costs in this workspace.
Yes. Choose Flare or Sunburst and the workspace sends your request to that model through EvoLink. It displays the finished image and a download link.
Flare is the default for everyday creation. Sunburst is intended for detailed work that prioritizes editing precision and may take longer. Compare the variants on the same brief when precision matters.
Yes. Upload up to 16 references and describe your changes. Files can be PNG, JPEG or WebP, up to 10 MB each; the workspace resizes references to a maximum 1024px edge.
Upload a PNG with an alpha channel and the same original dimensions as your first reference. Transparent areas are regenerated. The workspace applies the same resizing rule to references and masks.
No. The 4K tier has an approximately 8.29-megapixel budget. Square output is 2880 × 2880; 16:9 output is 3840 × 2160. Extreme ratios may use fewer pixels.
The quote depends on resolution, quality, prompt length and reference count. Short 1K drafts start at 2 credits. The workspace shows the total before generation; a successful task uses that fixed site quote.
Confirmed failed or canceled tasks receive an automatic credit refund. A network interruption is not proof of failure: keep the saved task and check its status before submitting another request.
The completed image is saved to site storage and appears in your recent tasks. Download PNG, JPEG or WebP according to the output format selected before generation.
Start with a 1K draft, refine your brief or reference images, then choose a higher quality setting for the next version.