Flare / Sunburst
Sunburst focuses on edit fidelity. Use it when preserving the reference matters most.
SceneFlareGPT Image 2.5
More image models
Explore other models for your next image generation or editing task.
GPT Image 2.5
Create a new image. Refine the details that matter.
Start from a written brief or work from your own photos. GPT Image 2.5 combines image generation and reference editing in one workspace, with Flare for faster iteration and Sunburst for edits where fidelity matters. Choose the version for each task without switching pages.
Flare / Sunburst1K / 2K / 4KEdit a region with a mask
Low → Maximum
Five quality levels let you balance detail, generation time and credits.
1–10
Choose 1–10 images per request. After completion, select each thumbnail to preview and download that image. Your results remain accessible from Library after leaving the page.
From a first draft to a focused edit
Generate and edit images with GPT Image 2.5. Choose Flare or Sunburst, add references and masks, set dimensions, and review credits before creating up to 10 images.
Build the scene from a clear brief
Describe the subject, setting, lighting and composition. For a poster or interface concept, put the exact words in quotation marks and specify their placement. Review the result for spelling, anatomy and product accuracy.
Guide changes with your reference images
Add up to 16 photos, sketches or layout references. Explain what each reference contributes and name what should remain unchanged. A reference guides the result; it does not guarantee an exact match.
Edit a region with a mask
Upload a PNG matching the first reference image’s dimensions. Transparent pixels mark the area to change; opaque pixels mark the area to preserve.
Size mode
Width and height must be multiples of 16, with at most 3,840 px per side, 655,360–8,294,400 total pixels and an aspect ratio no wider than 3:1.
Transparent · Preview
Transparent output is a preview feature. Check the alpha channel and edges before using the image.
Compare up to 10 outputs
Choose 1–10 images per request. After completion, select each thumbnail to preview and download that image. Your results remain accessible from Library after leaving the page.
See the model at work
Published examples from EvoLink and OpenAI illustrate scene generation and reference editing. They are source demonstrations, not SceneFlare benchmark results.
A dense city scene with a consistent visual direction
EvoLink’s Flare sample combines a neon city, people, architecture and night lighting. Use a similarly structured brief to establish the scene first, then refine individual details.
Flare
Flare is the faster choice for everyday generation and editing.
Change the clothing while keeping the portrait recognizable
OpenAI’s published comparison changes a red shirt into formalwear. Move the divider to compare the source and edited portrait. This demonstrates Images 2.5; the example does not identify a specific API subversion.
Guide changes with your reference images
Add up to 16 photos, sketches or layout references. Explain what each reference contributes and name what should remain unchanged. A reference guides the result; it does not guarantee an exact match.
How to use GPT Image 2.5
01
Choose a version and input mode
Start with Flare for iteration or Sunburst for precise edits. Select text-to-image for a new scene, or image-to-image to upload references.
02
Set the output and check the quote
Choose quality, size, background and image count. For local edits, add an alpha mask matching the first reference. The button shows the total credits before submission.
03
Generate, compare and download
Wait for completion, inspect every result and download the ones you need. To refine a result, upload it as the reference for a new request.
GPT Image 2.5 · FAQ
What is the difference between Flare and Sunburst?
Flare is the faster choice for everyday generation and editing. Sunburst focuses on edit fidelity. Use it when preserving the reference matters most.
Can I edit only part of an image?
Upload a PNG matching the first reference image’s dimensions. Transparent pixels mark the area to change; opaque pixels mark the area to preserve.
What does 4K mean on this page?
Output above 3,686,400 pixels is experimental. The 4K label is a pixel budget: 16:9 reaches 3840×2160; square output is 2880×2880.
Can I generate several images at once?
Choose 1–10 images per request. After completion, select each thumbnail to preview and download that image. Your results remain accessible from Library after leaving the page.
How are SceneFlare credits calculated?
Both versions use the same SceneFlare tariff. Medium quality at 1K starts at 6 credits for one image, including the first 1,024 UTF-8 bytes of prompt text. Each reference or mask adds 6 credits per output; each extra 1,024-byte prompt block adds 2. Multiply by the number of outputs. Auto uses the largest size band. The generation button is the final quote for your settings.
Is this the same as ChatGPT Images?
This page provides image generation and editing through EvoLink. ChatGPT-specific Sketch, image comments and conversational history are not part of this API workflow. You can supply your own sketch as a reference or upload a downloaded result for another edit.
Make your next image with GPT Image 2.5
Start with a prompt, choose your version and review the credits before generating.
Flare / Sunburst Version1–10 Images