GPT Image | v2.5 | Sunburst | Text to Image image preview

Example inputhover

size: "1024x1024"
prompt: "A photorealistic studio portrait of a young woman completely covered in sticky notes, plastered across her face, her hair, her shoulders and her blazer. She is beaming, a wide open genuine smile with her eyes crinkled, clearly delighted rather than overwhelmed. Short dark bob with a soft fringe, freckles, natural skin texture, wearing a beige oversized blazer over a white collared shirt and a slim dark tie. The sticky notes are orange, yellow, pink and pale green, layered densely and overlapping at slightly different angles, some curling at the corners. Every single orange note has "each::labs" written on it in neat handwritten black marker, clearly legible. The notes in the other colours have illegible scribbled handwriting and small doodles, no readable words. Plain warm grey seamless studio background. Soft even frontal lighting with gentle shadows under the notes. Shot on 85mm, shallow depth of field, sharp focus on her face, editorial fashion photography, fine grain."
quality: "auto"
background: "auto"
moderation: "low"
num_images: 1
output_format: "png"

GPT Image | v2.5 | Sunburst | Text to Image

Image Gen·gpt-image·by OpenAI

OpenAI’s high-fidelity image model, designed for polished visual output and fine-grained detail, with longer generation times in return for higher precision and quality.

Try it now →

API reference

Runtime (p50): -
Estimated price: From $0.053

Call the API

prediction.sh

curl -X POST \
  -H "Authorization: Bearer $EACHLABS_API_KEY" \
  -H "Content-Type: application/json" \
  --data '{
    "model": "gpt-image-v2-5-sunburst-text-to-image",
    "input": {
        "size": "1024x1024",
        "prompt": "A photorealistic studio portrait of a young woman completely covered in sticky notes, plastered across her face, her hair, her shoulders and her blazer. She is beaming, a wide open genuine smile with her eyes crinkled, clearly delighted rather than overwhelmed. Short dark bob with a soft fringe, freckles, natural skin texture, wearing a beige oversized blazer over a white collared shirt and a slim dark tie.\n\nThe sticky notes are orange, yellow, pink and pale green, layered densely and overlapping at slightly different angles, some curling at the corners. Every single orange note has \"each::labs\" written on it in neat handwritten black marker, clearly legible. The notes in the other colours have illegible scribbled handwriting and small doodles, no readable words.\n\nPlain warm grey seamless studio background. Soft even frontal lighting with gentle shadows under the notes. Shot on 85mm, shallow depth of field, sharp focus on her face, editorial fashion photography, fine grain.",
        "quality": "auto",
        "background": "auto",
        "moderation": "low",
        "num_images": 1,
        "output_format": "png"
    },
    "webhook_url": ""
}' \
  https://api.eachlabs.ai/v1/prediction/

Documentation8 sections

Overview
GPT Image | v2.5 | Sunburst | Text to Image Overview

GPT Image | v2.5 | Sunburst | Text to Image is OpenAI’s high-precision text-to-image model, designed for premium visual workflows that demand tight control over edits and fine-grained detail. Operating within the GPT Image 2.5 family, it extends the capabilities of earlier GPT Image 2 models by focusing on higher fidelity, more accurate instruction-following, and improved editing behavior. The primary differentiator of GPT Image | v2.5 | Sunburst | Text to Image is its ability to deliver polished, production-ready imagery at the cost of longer generation times compared to the faster Flare variant, making it ideal for campaign creative, product photography, and other detail-sensitive use cases.
Capabilities
Capabilities
- Generates high-fidelity images from natural language prompts, with stronger instruction-following and text accuracy than earlier GPT Image models.
- Performs precise image editing, allowing users to modify composition, lighting, background, or style while preserving core subjects across iterations.
- Supports workflows that combine text prompts with reference images, leveraging image inputs for guided edits and multi-image compositions.
- Offers multiple quality tiers—from low through max—so teams can prototype quickly at lower cost and finalize at premium quality.
- Integrates with the OpenAI text-to-image stack via the GPT Image | v2.5 | Sunburst | Text to Image API, using an OpenAI-compatible JSON format and image URL responses.
- Delivers improved subject preservation and editing consistency across multi-turn or multi-step creative workflows, reducing drift in complex projects.
Use cases
Use Cases for GPT Image | v2.5 | Sunburst | Text to Image

For brand and product designers, GPT Image | v2.5 | Sunburst | Text to Image excels at polished hero imagery, using its high-quality and precise editing to refine materials, lighting, and typography for storefronts and packaging. A typical prompt might be: “A premium skincare product hero shot on a reflective glass surface, soft diffused studio lighting, clean sans-serif logo text ‘LUMEA’ on the bottle, max quality.” Marketers can generate campaign-ready visuals where exact scene composition and on-brand colors matter, relying on Sunburst’s subject preservation to iterate safely on backgrounds and props without losing the core concept. Example: “Lifestyle scene of a young professional working in a sunlit home office, muted pastel color palette, brand accent color teal, high quality.” Developers integrating each::labs can expose GPT Image | v2.5 | Sunburst | Text to Image for premium tiers in their apps, using lower quality settings for previews and max quality for export-ready outputs via the GPT Image | v2.5 | Sunburst | Text to Image API. Creators and illustrators can use Sunburst to lock composition and then perform successive edits—such as changing time of day, wardrobe, or props—by modifying one variable per prompt for consistent series artwork.
Tips & tricks
Tips and Tricks

Effective prompting for GPT Image | v2.5 | Sunburst | Text to Image follows the same structured approach recommended for OpenAI text-to-image models like GPT Image 2: clearly describe subject, composition, action, setting, lighting, and style, then add constraints such as lens, aspect ratio, and resolution on a separate line. Keep text instructions short and explicit, and quote any embedded text you want to appear in the image, as GPT Image models render typography most reliably when the exact characters are written in quotes. Avoid conflicting instructions (for example, “no text” alongside a specific word) and limit style modifiers to one or two strong anchors to prevent washed-out results. For iterative workflows, change only one attribute per prompt—such as lighting or background—so Sunburst can preserve composition across edits. Example prompts include: “A studio product shot of a matte ceramic water bottle on warm linen, soft window light, neutral background, photorealistic, aspect ratio 3:2.”; “Editorial fashion portrait, subject mid-stride in a city street at golden hour, shallow depth of field, shot on 85mm, high quality.”; “Isometric 3D render of a modern office desk setup with a laptop and coffee mug, soft ambient lighting, clean minimalist style, 2K resolution.”
Technical spec
Technical Specifications
- Model family and ID: Part of OpenAI’s GPT Image 2.5 series; invoked via the gpt-image-2.5-sunburst model in the OpenAI Image API and Responses API image tool.
- Quality settings: Supports configurable quality levels low, medium, high, xhigh, max, and auto to trade off speed vs detail.
- Input formats: Text prompts and optional image inputs for editing and variations; inputs are billed as image and text tokens under OpenAI’s Image pricing.
- Output format: Generated images returned as URLs or binary image data via the OpenAI-compatible Image API response shape.
- Resolution and aspect ratio: Uses an aspect-ratio–driven workflow similar to GPT Image 2, supporting multiple resolutions from native 1K and 2K up to higher tiers, selected via aspect ratio and quality rather than a single fixed size.
- Latency profile: Sunburst is optimized for quality, with intentionally longer generation times than GPT Image 2.5 Flare in exchange for more precise control and detail.
Things to be aware of
Things to Be Aware Of

Because GPT Image | v2.5 | Sunburst | Text to Image is optimized for quality, generation times are longer than for Flare and comparable fast models, so synchronous calls may feel slower in interactive apps. Image generation and editing are subject to OpenAI’s separate image allowances and rate limits, which are distinct from text chat limits and can vary by plan and over time. High and max quality tiers consume more compute and cost, so using them for every iteration can quickly increase spend; many teams prototype at medium or high and reserve xhigh or max for final renders. As with other OpenAI text-to-image models, vague or overloaded prompts—too many styles, conflicting instructions, or long unstructured descriptions—can lead to inconsistent results, so structured prompt engineering is critical.
Key considerations
Key Considerations

GPT Image | v2.5 | Sunburst | Text to Image is best chosen when visual quality and edit precision are more important than raw speed, such as for polished product imagery or campaign assets. You should expect higher latency and plan workflows around synchronous, quality-configurable calls via the GPT Image | v2.5 | Sunburst | Text to Image API rather than rapid bulk generation. Both text and image inputs accrue token-based costs, so iterative editing workflows benefit from careful prompt design and selective use of high or max quality only for final renders. For fast drafts or high-volume social content, OpenAI text-to-image users may prefer starting with the Flare variant and reserving Sunburst for final passes where premium detail matters most.
Limitations
Limitations

GPT Image | v2.5 | Sunburst | Text to Image does not prioritize speed and may be less suitable for ultra-high-volume or real-time generation compared with smaller, faster text-to-image models. It remains bound by OpenAI’s safety systems and training distribution, so highly specialized domains, extremely niche aesthetics, or sensitive content may not render as expected. Users cannot bypass image-generation quotas or rate limits, and the GPT Image | v2.5 | Sunburst | Text to Image API returns images rather than raw text, making it a poor fit where downstream workflows depend on token-level textual output. Finally, although Sunburst improves text rendering and edit precision over GPT Image 2, some complex layouts or dense typography may still require manual design refinement.

Related models

4 models

Ideogram V4 | Text to Image AI model preview

Ideogram V4 | Text to ImageIdeogram

XAI | Grok | Imagine | 2.0 | Text to Image AI model preview

XAI | Grok | Imagine | 2.0 | Text to ImagexAI

Bytedance | Seedream | v5 | Pro | Text to Image AI model preview

Bytedance | Seedream | v5 | Pro | Text to ImageBytedance

Nano Banana 2 | Lite | Text to Image AI model preview

Nano Banana 2 | Lite | Text to ImageGoogle

GPT Image | v2.5 | Sunburst | Text to Image

GPT Image | v2.5 | Sunburst | Text to Image Overview

Capabilities

Use Cases for GPT Image | v2.5 | Sunburst | Text to Image

Tips and Tricks

Technical Specifications

Things to Be Aware Of

Key Considerations

Limitations

Related models