GPT Image 2.5 Sunburst model guide
A practical guide to GPT Image 2.5 Sunburst on GPT Image Hub: OpenAI's precision variant of GPT Image 2.5 for detailed creative work, what it shares with Flare, measured timings and output tokens, and when the extra time is worth it.
GPT Image 2.5 Sunburst is the GPT Image 2.5 model OpenAI recommends when precision matters most. It supports the same xhigh and max quality settings, custom resolutions, and transparent backgrounds as Flare, and GPT Image Hub runs it at xhigh by default; in Hub tests it took 47–108 seconds per image and placed every label and line of text exactly as prompted. Choose it for packaging, typography-heavy layouts and intricate illustration; choose Flare when speed matters more.
Visual examples
Visual examples
A packaging render edited so only the front label changes.
Label replacement
A packaging render edited so only the front label changes.
Replace the front label with the attached design at exact scale.Labeled illustration
A detailed botanical cross-section with short captions.
1024x1536 fig cross-section with five hand-lettered labels.Final render
One max-quality pass after approval.
Render the approved edit at max quality.Best use cases
- Masked or multi-reference edits where small details must land exactly where the prompt says.
- Detailed creative work such as packaging, typography-heavy layouts, and intricate illustration.
- Final renders at xhigh or max once a composition has been approved on a faster model.
Where to be careful
- High-volume drafting where latency matters; OpenAI positions Flare as the faster default.
- Resolutions above 2560×1440, which OpenAI labels experimental for the 2.5 models.
Strengths
OpenAI's recommended GPT Image 2.5 model for workflows where editing precision matters most.
Same quality ladder as Flare: low, medium, high, xhigh, max, and auto.
Same custom-size rules: multiples of 16, 1:3 to 3:1, no edge above 3840 px, 655,360 to 8,294,400 pixels in total.
Transparent backgrounds with PNG or WebP output, and the same per-token pricing as GPT Image 2.
Use cases
Precision edits
Label swaps, object replacement, and masked changes that must respect the rest of the image.
Detailed illustration
Intricate layouts, typography, and diagrams where small errors are expensive.
Approved finals
One careful xhigh or max render after the composition is settled elsewhere.
Workflow
Approve on a faster model
Settle composition and copy with GPT Image 2 or Flare before switching to Sunburst.
Write the edit as constraints
List what stays fixed, what changes, and where the mask applies.
Render once at high quality
Use xhigh or max for the final file and keep the output format consistent with the brief.
Prompting guidance
- Describe edits as explicit preserve-and-change instructions; precision is the reason to pick this model.
- When using a mask, keep it the same format and size as the source image with an alpha channel; the mask applies to the first input image.
- Use xhigh or max only for the final pass, since those settings produce more billed output tokens.
- Quote any on-image text exactly and keep it short.
Copy-ready prompts
Edit the reference packaging render: keep the box geometry and lighting, replace the front-panel label with the attached label design at exact scale, no other changes.
Create a 1024x1536 editorial illustration of a botanical cross-section of a fig, labeled with five short hand-lettered captions, muted sage and plum palette, paper texture.
Using the two reference photos, generate a precise before-and-after pair of the same living room with the sofa recolored to ochre and nothing else changed.
Example outputs generated on GPT Image Hub
Official sources
FAQ
Is GPT Image 2.5 Sunburst available on GPT Image Hub?
Yes. GPT Image 2.5 Sunburst is selectable in the GPT Image Hub generator with 1K, 2K and 4K tiers, and the Hub sends OpenAI's xhigh quality setting by default. Its credit cost per image is listed on the pricing page.
When should I pick Sunburst over Flare?
Pick Sunburst for edits and detailed creative work where precision matters most, and accept longer generation times. Pick Flare for fast, high-quality everyday generation.
Does Sunburst support the same sizes and quality settings as Flare?
Yes. OpenAI's image generation guide lists the same quality settings (low, medium, high, xhigh, max, auto) and the same custom-size constraints for both GPT Image 2.5 models.
Use this model
Start from a prompt template or compare models first
Use the generator when you already know the model, or compare all supported models when the prompt needs a different balance of speed, detail, and reference inputs.