Images · Text → image · Image → image
GPT Image 2.5
Sharp detail, natural light, and precise edits — in fast Flare and premium Sunburst
About the model
GPT Image 2.5 is OpenAI’s model update released in September 2026. Compared with GPT Image 2, it’s faster, delivers sharper detail and more natural lighting and textures, and — most importantly — treats the source image with more care: objects and recognizable details from a reference stay intact even as the environment, style, or composition changes.
The model comes in two versions. Flare is the default choice for most tasks: fast, with 2.5-level quality, and well suited to social content, promo graphics, and frequent iterations. Sunburst is built for premium work where precision and a polished result matter: final ad assets, brand visuals, product imagery.
In our service, both versions work in text-to-image and image-to-image modes. In the second mode, you can upload up to 16 references and precisely change the background, color, material, text, or a single object — the rest of the frame stays as it was. Resolution is 1K, 2K, or 4K.
Strengths
- Sharper detail, more natural lighting and textures than GPT Image 2
- Preserves objects, places, and recognizable details from references more faithfully
- Local edits to the background, color, material, or text without changing the rest of the frame
- Consistent results across successive edits of the same image
- Reliably follows complex prompts with multiple objects, text, and layout requirements
- Up to 16 references per request and resolution up to 4K
Best for
- Social graphics, covers, and promo images (Flare)
- Final ad assets and brand visuals (Sunburst)
- Product scenes and new environments for existing product shots
- Posters, banners, and landing-page graphics with a defined composition
- Rapid exploration of concepts and layout variants
Limitations
- Maximum of 16 input images per request
- Each file up to 30 MB
- Supported formats: JPEG, PNG, WEBP
- Resolution: 1K, 2K, or 4K
Prompting tips
- Flare is the default for most tasks; pick Sunburst when maximum polish and control matter
- When editing, start with what must stay unchanged, then describe the edit
- For a series of edits, upload the latest result as a reference and change one element at a time
- Put in-image text in quotes and specify the language
- In complex prompts, list the objects, their placement, and the style separately