GPT Image 2.5 Sunburst

GPT Image 2.5 Sunburst is OpenAI's highest-capability, precision-oriented image generation and editing model in the GPT Image 2.5 family. Its official API ID is gpt-image-2.5-sunburst. It accepts text and image inputs and returns images, supporting creation from prompts as well as composition, redesign, and localized edits based on one or more references. The dated gpt-image-2.5-sunburst-2026-09-08 snapshot is available for version-pinned workflows.

The model works directly through the Image API and as the image_generation tool in the Responses API. It supports quality settings from low through max, custom canvas dimensions, PNG/JPEG/WebP output, transparent backgrounds, compression controls, and streaming partial previews. It is designed for creative tasks where reference retention, editing precision, and delivery specifications matter most.

Key Capabilities

• Precision Text-to-Image: Converts detailed prompts into photography, illustration, posters, interface concepts, and other visual assets.
• Complex Image Editing: Changes subjects, environments, composition, lighting, materials, style, or local details through instructions.
• Multi-Image Reference Composition: Extracts products, people, or visual elements from multiple inputs and combines them into one scene.
• Mask-Guided Editing: Uses a mask to indicate the preferred edit area while the prompt defines retained and changed elements.
• Advanced Quality Control: Provides low, medium, high, xhigh, max, and auto quality settings.
• Flexible Canvas and Transparency: Supports standard and custom dimensions plus transparent-background design assets.
• Multi-Turn Visual Iteration: Uses Responses API context to continue refining the direction and details of generated imagery.

Technical Strengths

Feature	Benefit
Precision-Oriented Family Tier	Serves complex editing and demanding reference-image workflows as the detailed-production option in the 2.5 family.
Two API Paths	The Image API handles direct creation and editing, while the Responses API supports contextual visual workflows.
Custom Dimension Support	Accepts dimensions in 16-pixel increments within documented aspect-ratio and total-pixel constraints.
Complete Output Controls	Configures size, quality, format, JPEG/WebP compression, and transparent, opaque, or automatic backgrounds.
Streaming Partial Images	Returns up to three partial previews during generation for more responsive interactive products.
Pinnable Model Snapshot	The dated version provides a stable evaluation baseline when the continuously updated ID would be unsuitable.

Frequently Asked Questions

How does GPT Image 2.5 Sunburst differ from GPT Image 2.5 Flare?

GPT Image 2.5 Sunburst targets maximum capability and editing precision, while Flare targets fast, high-quality everyday image generation. Start with Sunburst for complex reference composition, local replacement, or precision-sensitive revisions, and evaluate Flare for rapid creative exploration at scale.

Can GPT Image 2.5 Sunburst use multiple reference images?

Yes, GPT Image 2.5 Sunburst can use multiple image inputs for composition and editing. Define the role of each image, the attributes to preserve, and the final spatial relationship in the prompt so similar people, products, or textures are less likely to be confused.

Is GPT Image 2.5 Sunburst mask editing pixel-perfect?

No, GPT Image 2.5 Sunburst treats a mask as prompt-based guidance rather than a hard pixel boundary. With multiple input images, the mask applies only to the first image, so that image should be the base canvas and the prompt should explicitly identify everything that must remain unchanged.

How does GPT Image 2.5 Sunburst create transparent backgrounds?

Set background to transparent and choose PNG or WebP output for a GPT Image 2.5 Sunburst transparent-background request. JPEG cannot retain alpha transparency, and the prompt should describe edge treatment, spacing, contact shadows, and whether the full subject outline must remain visible.

How should GPT Image 2.5 Sunburst support be verified after it appears on LinkModel?

Open the live LinkModel model page and confirm its platform ID, generation and editing entry points, reference-image count, mask field, custom dimensions, quality settings, and transparent-background support. Integrate only against LinkModel's published request schema rather than copying OpenAI-native URLs or Responses API tool structures.