AI image model · OpenAI
GPT Image 2.5 Flare, free at 1K.
GPT Image 2.5 Flare is OpenAI’s fast everyday model, and the first in this catalog with a genuine 4K at every shape — the square included. Quick enough to keep changing your mind, cheap enough that the opening tier is on the free plan.
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 Flare.
GPT Image 2.5 Flare vs GPT Image 2.5 Sunburst
Two models, different jobs.
Flare is the one you start on. Both take the same sizes and cost the same credits, so switching is only ever a trade of speed for precision.
GPT Image 2.5 Flare
GPT Image 2.5 Sunburst
Best at
Everyday work and iteration
Precise edits and final assets
Speed
Very fast
Slower, more deliberate
Reach for it when
You want several directions before committing to one
A face, a product or small text has to survive the edit
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 Sunburst is for.
Why this model
What GPT Image 2.5 Flare is for.
Fast enough to iterate
OpenAI’s quickest model at this quality. Most sizes come back inside half a minute, so you can run an idea, look at it and rewrite the prompt without losing the thread.
A 4K that is actually 4K
Square, wide and tall all render at full size: 2880x2880, 3840x2160, 2160x3840. Nothing else here draws a true 4K square, and it costs 4 credits.
The GPT way with instructions
It takes direction the way the GPT family always has — composition, framing, what to leave untouched — and reads up to eight reference images in Custom mode, four on the free plan.
Three tiers, three real sizes
1K for 2 credits, 2K for 3, 4K for 4. Every rung is a different output size, never a quality flag on the same pixels.
Prompt guide
How to prompt GPT Image 2.5 Flare.
Flare rewards a short prompt run many times, not a long one run once. The trick is a style block you freeze and a single thing you vary.
The shape
Intent: what the image is for
Hero subject: the subject, and where it sits in the frame
Finishing details: style, palette, what to leave out
Camera: angle and framingA worked example
Intent: editorial cover art for an article
about prompt engineering.
Hero subject: one clean conceptual object,
centred, the right third left empty for a
headline to be added later.
Finishing details: flat modern illustration,
warm limited palette, thin ink-blue linework,
no gradients, no text, no logos.
Camera: straight-on, flat layout.Freeze the style block
When you want a set, change one word and nothing else. Rewriting the description between items is what makes six thumbnails look like six different projects.
Start at the opening tier
Run 1K first and step up only when a real requirement is unmet. A higher tier buys pixels, not a better idea, and it does not rescue a vague prompt.
Leave the words to the layout
Ask for negative space instead of a headline. Text set in your design tool stays editable and always renders correctly, which is the right division of labour rather than a limitation.
Escalate, don’t grind
If the same prompt misses twice for a reason of precision — a face must survive, small type must be perfect — that is a Sunburst job. Re-rolling on Flare will not get there.
Use cases
Where GPT Image 2.5 Flare fits.
Social and thumbnails
Feed posts, story backgrounds and gallery cards, at the cadence a content calendar actually runs on.
Exploring a direction
Render the same subject four ways, pick the winner, then rebuild it properly. Cheap enough that being wrong three times costs nothing.
In-product art
Empty states, placeholder avatars and surface patterns, on a real alpha channel where you need one.
Pre-visualisation
A rough sketch of the idea before you spend a slower render on 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.
Can I use GPT Image 2.5 Flare without a ChatGPT subscription?
Yes. ReelyArt runs it server-side: a free account generates at 1K (2 credits) with 10 signup credits, no OpenAI account, API key or ChatGPT plan involved.
Flare or Sunburst?
Flare for everyday work and iteration, Sunburst when an edit has to land exactly. They cost the same here and take the same sizes; what differs is speed against precision.
Is the 4K tier a real 4K?
At all five ratios. OpenAI caps a request at 8.3 million pixels, and these are the largest sizes that fit under it: 2880x2880 for the square, 3840x2160 and 2160x3840 for wide and tall, 3312x2480 and 2480x3312 for 4:3.
How many reference images can I attach?
Eight on a paid plan, four on the free plan. Most Flare work needs one or none — it is the model you reach for when you are describing something, not rebuilding it.
What does one image cost?
2 credits at 1K, 3 at 2K, 4 at 4K — and four variations cost four times that, which is the usual way to spend on this model. Failed generations are refunded automatically.