Powered by ChatGPT Images 2.5, OpenAI's newest image model
GPT Images turns ChatGPT Images 2.5 into a real workspace for making images, not just generating them. Sketch to lock the composition, pin edits on the picture, blend up to 16 reference photos — every version kept, every edit reversible.
Free credits on sign-up · No credit card required
This is the signature ChatGPT Images 2.5 workflow: stick figures and boxes are enough to fix the layout — where the subject stands, how large, where the horizon sits. Add one line of direction and the model renders the rest.
Drag the handle to compare
The full ChatGPT Images 2.5 workspace, embedded on this page. Describe, sketch, or mark up a photo — you only need an account when you hit Generate.
Preview
Your generated image appears here. Describe it, attach a sketch, or mark up a reference photo on the left.
Six capabilities that change how images get made — wired into a canvas built for iteration.
The model reads the geometry, perspective and proportions of your drawing, then renders a finished image that follows it. Composition and framing become things you show, not describe — a stick figure on a rooftop really can become a movie still.
Click anywhere on a generated image, drop a numbered pin, and write what should change: "make this sign read OPEN", "remove the headphones". Only the marked regions are repainted, every other pixel stays put — several pins are handled in one pass.
The old pain of AI editing — ask for a small change, get a different person — is gone. Faces, outfits, product geometry and logos stay anchored across rounds, so version two is version one plus your edit, nothing more.
Blend people, products and places from separate photos into one coherent scene. Two award winners from different cities become a believable joint celebration; a phone snapshot becomes a studio-lit hero image, identity intact.
ChatGPT Images 2.5 ships as Flare and Sunburst. Flare answers in 5–10 seconds for drafts and batches; Sunburst thinks for 25–45 seconds and delivers print-grade detail. Switch per generation — same workspace, same credits.
Posters, covers and infographics come out with legible headlines instead of alphabet soup. Spelling accuracy is dramatically improved, so flyers and product cards arrive ready to use — though dense copy still deserves a proofread.
Three ways to tell the model what you want. Use whichever fits the thought — they share one canvas.
Write plain language: a mood, a scene, a style. Templates give you proven prompt scaffolds for posters, covers and product shots, so you never start from a blank box.
Open the built-in canvas and draw the layout: a person here, buildings there. When position and framing are hard to phrase, five seconds of drawing beats two hundred words.
Already 90% happy? Don't regenerate — point. Click the spot that bothers you and say what should happen there. Local edits that respect everything you already love.
ChatGPT Images 2.5 splits its API into two models on one pricing table. GPT Images routes each request to the right one — fast drafts on Flare, meticulous finals on Sunburst — and shows the exact credit cost before you click generate.
Everyday speed
About 50% lower latency than the previous generation. Ideal for social posts, e-commerce batches and any workflow where you iterate quickly.
Editing precision
Deeper multi-step reasoning for complex compositions, layered typography and micro-texture — for 4K ads, print posters and fine multi-region edits.
| Quality | Output tokens | API cost | Credits here |
|---|---|---|---|
| low | 196 | $0.006 | 1 |
| medium | 439 | $0.013 | 2 |
| high | 1,756 | $0.053 | 6 |
| xhigh | 3,122 | $0.094 | 10 |
| max | 7,024 | $0.211 | 22 |
Token rates are identical on both models: $30 per 1M output tokens, $8 per 1M image input tokens. GPT Images converts this to simple credits — no token math required of you.
Real workflows, verified in hands-on reviews — not demo-day cherry picks.
A phone snapshot becomes a polished product poster, then a full detail-page set: selling points, close-ups, scenes, spec cards — minutes, not half a day.
Article covers and thumbnails with your actual face, edited round after round — move right, clear space for the title — without identity ever drifting.
Decades-old family photos repaired in one pass: creases and stains removed, detail rebuilt, natural color added — while Grandpa still looks like Grandpa.
Two people, two cities, one trophy each — blended into a single believable celebration shot with matched lighting, perspective and proportions.
Sketch a shot — subject right, city left, billboard far — and get a cinematic keyframe for your video model. Script to storyboard to keyframe, no prompt gymnastics.
Answer a few questions, get logo directions, then see them on tote bags, mugs, T-shirts and stickers. Mockup hunting is officially over.
Every generation shows its credit cost upfront, computed from real ChatGPT Images 2.5 API rates. Failed generations refund automatically. Start free, top up when hooked, subscribe when shipping daily.
For weekly creators
$6.93/mo
$9.90/mo
$83.16/yr
$118.80/yr
For daily professionals
$20.93/mo
$29.90/mo
$251.16/yr
$358.80/yr
For serious creators
$55.93/mo
$79.90/mo
$671.16/yr
$958.80/yr
100 credits, never expire
$5 · one-time
550 credits, never expire
$25 · one-time
2400 credits, never expire
$99 · one-time
What a generation costs
Draft (Flare · medium): 2 credits
Standard (high): 6 credits
Masterpiece (Sunburst · max, 4K): 88 credits
Everything you'd want to know before your first render.
Join creators already shipping posters, covers and product visuals. Free credits on sign-up — no card, no install, no prompt engineering required.
