Powered by OpenAI GPT Image 2.5 — Flare & Sunburst. Describe an image and GPT Image 2.5 makes it. Add a reference photo and it edits only what you asked for — the subject, lighting and composition around it stay put.
DISCLAIMERimages25.com runs OpenAI’s GPT Image 2.5. We are an independent product and not affiliated with OpenAI. Sample images and video are OpenAI’s official ChatGPT Images 2.5 launch assets, shown for illustration.
Have a photo you want changed?
Try ChatGPT Images 2.5Sign in with Google · Plans from $8.33 a month, or buy credits once
Write a detailed prompt in natural language. Be specific about subjects, lighting, composition, type, and style. Think like a Creative Director.
Pick Flare or Sunburst, an aspect ratio, and 1K or 2K. Images25 runs OpenAI’s gpt-image-2.5 through a focused workspace — no chat tab-switching.
Upload a reference, describe the change, and download full-resolution files — we add no visible watermark of our own. Iterate in plain language instead of re-rolling from scratch.
Why The Edits Hold Up
ChatGPT Images 2.5 is OpenAI’s newest image model, released September 8, 2026. It renders light and texture more believably, holds onto the subject in your reference photo, and follows editing instructions across multiple turns — so the third revision still looks like the first.
Better preservation of the subjects in your reference photos — the same person, pet, or product comes out the other side still recognisable.
GPT Image 2.5 is better at editing only what you asked for, keeping the rest of the details the same — even on complex subjects and backgrounds.
Earlier changes stay consistent as the thread gets longer, and quality does not degrade the way it did over repeated 2.0 edits.
Better at reflecting a requested visual direction and holding composition together as the brief becomes more specific.
A photo of a printed studio portrait — the kind of imperfect source people actually have, not a clean studio file.
New outfit, same child, same pose, same blue background. That gap — changing what was asked and nothing else — is the 2.5 upgrade in one frame.
More believable light and shadow behaviour than ChatGPT Images 2.0 — the single most visible upgrade in this release.
Surface detail holds up under scrutiny: fabric, skin, metal, and print read as their own materials.
Reference photos keep their subject through generation and editing, instead of drifting into a lookalike.
Editing instructions are followed more reliably turn after turn, so a long revision thread stays coherent.
Generation latency cut by as much as half versus ChatGPT Images 2.0, carried into Flare here.
Flare for speed, Sunburst for precision — chosen per prompt on the same screen, no re-setup.
Handles more complex layouts, including transparent backgrounds — usable as a layer instead of a flat rectangle.
Change one product, background, or line of copy while the subject, composition, and brand treatment around it stay put.
Campaign visuals where the lighting has to look shot rather than generated — then reformatted for each placement.
Put a reference subject into new scenes and keep them recognisable across the whole set.
Product stills, lifestyle mockups, and packaging drafts where material texture is the difference between draft and usable.
Long revision threads — change the light, keep the subject, adjust the crop — without losing what already worked.
| GPT Image 2.5 Flare | GPT Image 2.5 Sunburst | GPT Image 2 | |
| Lighting & texture | 2.5 generation | 2.5 generation | Previous generation |
| Reference subject fidelity | Improved | Improved | Baseline |
| Multi-turn edit reliability | Improved | Improved | Baseline |
| Generation speed | Up to 50% faster | Longer by design | Baseline |
| Best for | Everyday iteration | Detailed creative work | — |
Lighting is where 2.5 pulls ahead, so spend your words there: “late afternoon window light from camera left, soft falloff, no fill.”
Upload the face, product, or object you need kept, then describe only the scene around it. Subject preservation is what this model is built for.
If an image is 80% right, say “keep the subject and framing, change the light to golden hour.” Instructions now hold across turns.
OpenAI’s default choice for most applications: higher-quality images than GPT Image 2 at 50% lower latency. Suited to creator and social content, product experiences, and rapid prototyping. Default on Images25.
Built for premium visual workflows that benefit from tighter control across edits — production-ready campaign creative or polished product imagery. Slower by design.
Sketch, Templates, and comment-on-image editing shipped as ChatGPT app features, not API capabilities — so they live in ChatGPT, not here. Images25 gives you the models themselves: generate, edit with a reference, and download at full resolution with no visible watermark of ours.
Credits are the only thing that changes between plans: same two models, same sizes, same licence. Pay by card, billed in USD.
For occasional projects and trying the model properly
$215/year billed yearly
40% OFF · Save $143.8
100 credits ≈ $2.24
POPULAR
For creators and marketers who generate every week
$360/year billed yearly
40% OFF · Save $238.8
100 credits ≈ $1.67
BEST VALUE
For studios and teams that render at volume
$599/year billed yearly
50% OFF · Save $599.8
100 credits ≈ $1.25
Every image costs credits, shown next to the generate button before you run it: a Flare image is 3 credits at 1K or 5 at 2K, a Sunburst image 10 at 1K or 20 at 2K. All prices in USD.
Generating at a volume none of these plans covers? Write to support@images25.com. Refunds: see our refund policy.
The September 8, 2026 launch, what Flare and Sunburst each do, what lives in ChatGPT versus the API, and how Images25 runs them.
ChatGPT Images 2.5 is OpenAI’s newest image model, released September 8, 2026. It produces more natural lighting and richer textures, is better at preserving the subjects in your reference photos, and follows editing instructions more reliably across multiple turns. OpenAI also cut generation latency by up to 50% compared with ChatGPT Images 2.0.
OpenAI’s newest image model, in one focused workspace. Pick a plan and start generating.
IMAGES25
