AI image model · OpenAI
GPT Image 2.5 Sunburst, for edits that have to land.
GPT Image 2.5 Sunburst is OpenAI’s most capable image model, and the one to reach for when a change has to happen exactly where you asked and nowhere else. Same sizes and same price as Flare; it simply takes longer and holds more.
4K
top resolution
2 credits
per image, from
8 images
reference cap, 4 on free
1K
on the free plan
Real outputs
Straight out of GPT Image 2.5 Sunburst.
GPT Image 2.5 Sunburst vs GPT Image 2.5 Flare
Two models, different jobs.
Sunburst is the one you finish on. Both take the same sizes and cost the same credits, so nothing is lost by drafting elsewhere and bringing the winner here.
GPT Image 2.5 Sunburst
GPT Image 2.5 Flare
Best at
Precise edits and final assets
Everyday work and iteration
Speed
Slower, more deliberate
Very fast
Reach for it when
A face, a product or small text has to survive the edit
You want several directions before committing to one
Cost here
2 / 3 / 4 credits
2 / 3 / 4 credits
Not sure? Start here and move over when a detail has to survive. See what GPT Image 2.5 Flare is for.
Why this model
What GPT Image 2.5 Sunburst is for.
It edits rather than redraws
Hand it a photo and a specific instruction and the parts you did not mention stay as they were. That is the whole reason to pick it over Flare.
Detail worth the extra seconds
Skin, fabric, metal, foliage: at 2K and 4K it resolves texture that the faster models smooth over, which is what makes it the one to finish on.
Eight references, one scene
Give it a set of product shots or a character from several angles and it composes them into a single image instead of collaging them.
Priced like Flare, not like a flagship
Token for token it bills what Flare does, so ReelyArt charges the same: 2 credits at 1K, 3 at 2K, 4 at 4K.
Prompt guide
How to prompt GPT Image 2.5 Sunburst.
Sunburst reads a prompt in slots. Naming them is what turns a wish into a brief, and it is what makes an edit land on the one thing you meant to change.
The shape
Intent: what the image is
Background: the environment
Foreground: the nearest details
Hero subject: the subject, and where it sits
Text: the exact words, quoted, if any
Finishing details: materials, light, what to leave out
Camera: angle and framingA worked example
Intent: product photography for a design shop.
Background: warm neutral studio gradient.
Foreground: a soft shadow on pale limestone.
Hero subject: a matte sand-coloured ceramic
caddy holding two pencils and a brass ruler,
centred slightly right of frame.
Text: "REELYART" and "No. 04" on the label,
exactly once each.
Finishing details: fine clay grain, crisp
brass reflections, no extra text.
Camera: 70mm, eye-level, soft light upper left.Quote the words you need
Put every string in quotation marks and say where it goes and how many times it appears. A described caption comes back paraphrased; a quoted one comes back as written.
For an edit, write the preserve list
Name what must not move — face, pose, lighting direction, camera angle, framing — then name the single thing that changes. Restate the list on every follow-up, because it is what stops an edit from becoming a redraw.
Give every reference a role
With more than one image, say which is the subject, which is the background and which is style only. An unlabelled style reference will import objects you never asked for.
Say where things belong
“Slightly left of centre”, “in the lower third”, “negative space on the right for copy”. Placement is the instruction it follows best, and the one most prompts leave out.
Direct, don’t decorate
Stacked adjectives — masterpiece, ultra-detailed, award-winning — buy nothing. Material, light, camera, composition and what the subject is doing buy everything.
Use cases
Where GPT Image 2.5 Sunburst fits.
Final campaign assets
The frame that ships: the one where a mis-set word or a drifting product shape is not a re-roll, it is a reprint.
Controlled edits
Change a background, a garment or a label while the face, the pose and the light stay exactly as they were.
Product imagery
Turn a reference photograph into studio work, then put the real label on it in a second pass.
Layouts with real copy
Posters, packaging and covers where the text has to be spelled correctly and sit where you put it.
Pricing
What one image costs.
Credits per image; four variations cost four times the number shown. Aspect ratios: 1:1 · 16:9 · 9:16 · 4:3 · 3:4. Failed generations are refunded automatically. Plans and credit packs are on the pricing page.
FAQ
Before you generate.
Is GPT Image 2.5 Sunburst free?
Its 1K tier is: 2 credits an image on the free plan, which the 10 signup credits cover 5 times over. 2K and 4K are on any paid plan.
What is Sunburst best at?
Editing a real photograph. When the brief is “change this, keep everything else”, it holds the everything-else better than anything else in the catalog.
Sunburst or Flare?
Same price, same sizes, same reference cap. Sunburst is slower and more precise; Flare is the faster everyday pick. Draft on Flare, re-run on Sunburst when the detail matters.
How many reference images can I give it?
Eight on a paid plan, four on the free plan. Give each one a role in the prompt — subject, background, style only — or an unlabelled reference will import objects you never asked for.
What does one image cost?
2 credits at 1K, 3 at 2K, 4 at 4K — the same as Flare, so a slower render never costs more. Failed generations are refunded automatically.