GPT Image 2.5 Flare: AI Image Editor Online
Bottom line
GPT Image 2.5 Flare editing: up to 16 reference images, 1K/2K/4K, same price as Sunburst (3, 5, or 8 credits). Open the Image Edit tool, select GPT Image 2.5 Flare, upload your image, and the exact credit cost appears before the edit runs.
OpenAI released GPT Image 2.5 on 2026-09-08 in two variants that KIE exposes as four separate models on VdoBloom — Flare and Sunburst, each with its own text-to-image and image-to-image id. Flare is the fast, default variant; Sunburst trades some of that speed for edit precision and fidelity. Both variants accept the exact same input schema and — this is the part worth knowing before picking one — cost exactly the same on VdoBloom. Sunburst is not a premium tier; it trades latency, not price.
On the editing side, Flare's headline spec is input capacity: up to 16 reference images per edit, up from GPT Image 2's 10 — useful for jobs like a multi-panel sprite sheet or a composite that genuinely needs to see many source images at once. It shares the same 13 aspect ratios as the text-to-image side, the same 1K/2K/4K resolution ladder priced at 3, 5, and 8 credits, and the same 20,000-character prompt budget.
Unlike GPT Image 2's editor, there's no 1:1-at-4K restriction here — the only resolution limit is on four exotic ratios (27:16, 16:27, 9:8, 8:9), which clamp to 1K automatically rather than failing the job.
It carries VdoBloom's STRICT content label and a 4.7 user rating. Credit cost is shown before every edit runs.
Flare vs Sunburst for editing
Identical price, identical 16-image and resolution limits — the difference is purely which OpenAI optimized for. Flare is the faster default; Sunburst (rated 4.8 vs Flare's 4.7 on VdoBloom) trades some speed for precision. Since the cost doesn't change, it costs nothing to try the same edit on both and keep the better result.
What it costs on VdoBloom
3/5/8 credits for 1K/2K/4K, matching GPT Image 2.5's text-to-image pricing. A 16-image composite edit at 2K runs 5 credits regardless of how many of the 16 reference slots you actually fill.
Common editing problems and how to avoid them
Reports of noisy or inconsistent output usually trace back to one of two things: an ambiguous instruction that left the model guessing, or a resolution silently clamped to 1K because the aspect ratio picked was one of the four 1K-only ratios. Writing the instruction as explicit, separated steps and checking the resolution actually applied before judging the result addresses most of what gets reported as "bad" output. A failed job is refunded automatically, so it's worth retrying with a tighter prompt rather than assuming the model can't do the edit.
What creators use GPT Image 2.5 Flare for
- Edits that genuinely need many source images at once — up to 16 — such as multi-frame sprite sheets or large composites.
- Instruction-heavy edits where a long, literal, multi-step prompt (up to 20,000 characters) needs to be followed precisely.
- Draft-then-finalize workflows: 1K attempts at 3 credits, then a 4K delivery pass at 8 credits.
How to use GPT Image 2.5 Flare on VdoBloom
- 1Open the Image Edit tool and select GPT Image 2.5 Flare.
- 2Upload the image(s) to edit and describe the change you want.
- 3Review the credit cost and generate — results appear in your library.
Frequently asked questions
GPT Image 2.5 Flare vs Sunburst for editing — which should I use?
Flare is faster; Sunburst trades speed for edit precision and fidelity. Both accept the same inputs and cost the same — 3/5/8 credits by resolution — so there's no price penalty for trying Sunburst on a job where precision matters more.
How many reference images can GPT Image 2.5 Flare take in one edit?
Up to 16 — six more than GPT Image 2's 10-image limit, and the same ceiling as Sunburst.
How much does GPT Image 2.5 Flare editing cost?
3 credits at 1K, 5 at 2K, 8 at 4K — identical to the text-to-image side and to Sunburst.
Is it free to try?
10 free signup credits cover three 1K edits, or one 2K edit with 5 credits left over.
How strict is content filtering on GPT Image 2.5 Flare edits?
STRICT — VdoBloom's tightest tier, matching OpenAI's provider-side filtering, the same as GPT Image 2 and GPT Image 1.5.
GPT Image 2.5 vs GPT Image 2 for editing — what changed?
Reference image capacity nearly doubled (16 vs 10), the 1:1-can't-be-4K restriction is gone, and four resolution-limited aspect ratios now clamp to 1K instead of rejecting the job outright. Pricing is unchanged.
Users report inconsistent quality or visible noise on some GPT Image 2.5 outputs — why?
That kind of variance is typically prompt- or setting-driven rather than a fixed flaw: ambiguous instructions, an aspect ratio that got silently clamped to 1K, or a job run at 1K when 2K/4K was needed all produce softer or noisier results. If a specific result looks wrong, try Sunburst (same price) for a more precision-focused pass, or bump the resolution explicitly.
Can GPT Image 2.5 Flare handle a multi-frame sprite sheet edit?
It's well suited to that kind of job — its 16-image input and long prompt budget let you specify frame-by-frame instructions in one request, though how well any individual sheet turns out still depends on prompt clarity.