GPT Image 2.5 AI Image Generator
GPT Image 2.5 is the image model OpenAI shipped on 8 September 2026, in two API tiers: Flare for speed and Sunburst for editing precision. Start free here with no sign-up, on GPT Image 2 — and this page runs 2.5 the day a provider lists it.
GPT Image 2.5 AI Image Generator
Made with GPT Image 2.5 by its first users
Posted publicly within a day of the 8 September launch, each by a creator who named the model outright. This is the model's range in other people's hands — the generator on this page runs GPT Image 2.
One character, nine frames
What is GPT Image 2.5?
GPT Image 2.5 is the image model OpenAI released on 8 September 2026 as ChatGPT Images 2.5, calling it their new state-of-the-art image model. It renders pictures from a written prompt and edits images you supply by description, and it reaches the API as two model ids rather than one, so the choice between speed and editing precision is made when you pick the model.
More than three billion images have been generated across ChatGPT Images and the GPT-Image models in the API, and GPT Image 2.5 is the version that inherits that. The model details here come from OpenAI's launch post; the examples and use cases also draw on creators who publicly named GPT Image 2.5. What the button below runs today is GPT Image 2 — the same family, same instruction-led editing, live right now — because no provider on our shelf has listed 2.5 yet. The day one does, this page runs it.
Creative engine
What runs when you press Generate
GPT Image 2, billed per image by output size, with the estimate shown before you commit. GPT Image 2.5 goes live here the day a provider carries it.
Flare is the fast one
OpenAI calls gpt-image-2.5-flare its fastest model for high-quality everyday image generation, delivering higher-quality images than GPT Image 2 at 50% lower latency, with early testing at two to four times the speed. Named territory: creator and social content, visual search, rapid prototyping, high-volume generation.
Sunburst is the precise one
gpt-image-2.5-sunburst is built for premium visual workflows that benefit from tighter control across edits — campaign creative, polished product imagery — and takes longer to generate in exchange.
Edits hold their ground
The headline change is editing behaviour. GPT Image 2.5 changes only the elements you asked about and leaves the rest intact, even with complex subjects and busy backgrounds, and follows instructions more reliably across several turns — which is the half that decides whether you stay out of a layer stack.
Your reference photo survives
OpenAI singles out subject preservation: people and objects in a reference photo carry through instead of drifting into someone who merely resembles them. The launch notes also claim more natural lighting, richer textures, and more accurate rendering of images containing real-world information — maps, charts, packaging, legible type.
The model this page runs today has its own page at GPT Image 2, and the same engine under its consumer name at ChatGPT Image Generator. Nano Banana 2 and Nano Banana Pro are the other house style; Qwen Image 3 spells best.
When GPT Image 2.5 is the one you want
Three jobs the launch material points at, and what you can do about each one today.
The edit has to stay put
Multi-turn edits that touch only what you named are Sunburst's whole pitch. The closest thing running here is instruction-led editing on GPT Image 2.
You are generating at volume
Flare targets creator and social content, visual search and rapid prototyping — where halved latency compounds across a hundred runs rather than saving a moment on one.
The frame carries words
Rendering real-world information accurately is called out in the launch notes, and posters, diagrams and packaging are where that shows first.
Features
The model layer comes from OpenAI's launch post; the runtime layer comes from this site's configuration.
Two API model ids
gpt-image-2.5-flare for speed, gpt-image-2.5-sunburst for editing precision. They share one price list, so the choice is how long you will wait, not budget.
Sketch as a reference
In ChatGPT you can draw directly and hand that drawing to the model as the reference for the final image.
Templates and comments
Templates give starting points for popular formats, comments can be placed on an image to aim an edit, and prompts can be shared.
Text-to-image and editing here
The generator here does text-to-image and description-led editing on GPT Image 2, with the credit estimate shown before you commit.
Failed runs are free
A generation that does not complete is refunded automatically, unless it broke the content policy.
GPT Image 2.5 specifications
From OpenAI's launch material and this site's runtime config. Rows marked not stated are not published yet.
- Released
- 8 September 2026, as ChatGPT Images 2.5
- API model ids
- gpt-image-2.5-flare, gpt-image-2.5-sunburst
- API pricing
- Image output $30.00 / 1M tokens; image input $8.00 ($2.00 cached); text input $5.00 ($1.25 cached)
- Latency, Flare
- 50% lower than GPT Image 2; two to four times the speed in early testing
- Resolutions and aspect ratios
- Not stated in the launch material
- Available on Kavel
- Not yet — this page runs GPT Image 2 until a provider lists 2.5
Generating on this page
Three steps. The generator runs GPT Image 2, and the last step is the only place cost is decided.
Pick generate or edit
Text-to-image invents a picture; image-to-image changes one you already have when you describe the change. No mask, no layers — the sentence is the interface.
Name the specifics
Subject, framing, light and material. Vague prompts return the most photographed version of the idea on every model in this class.
Choose the size last
Settle the picture at 1K, then re-run the keeper larger. Size sharpens the render; it does not rearrange the frame.
What people are making with GPT Image 2.5
Drawn from what its first users posted after launch, and from the workflows OpenAI names for each tier.
Campaign and product creative
The workflow Sunburst is named for: campaign visuals and product imagery where an edit must hold across rounds of feedback.
Diagrams and reference pages
Dense typographic layouts — equations, captions, legends — readable at the size they will actually be printed at.
Character sheets
One figure across many poses, where subject preservation is either there or obviously missing.
Social and creator volume
Flare's stated territory: everyday images at speed, where waiting half as long changes what you attempt.
Retouching by description
Change a colour, drop a distraction, swap a garment — by saying so instead of masking it. This already works on the model this page runs.
Kavel pricing — you pay per generation, never per seat
One credit pool covers Nano Banana images and Seedance and Veo video. The cost shows before every run, the safety check runs before the model does, and a failed run is refunded automatically. Use a subscription for ongoing work, or a one-time pack when you just need to top up.
Free
What's included
- 15 credits before you sign up
- +40 when you create an account
- +100 a week from daily check-ins
- = 488 a month — 97 images or 12 videos
- No card, no account
- Free engine every dayGrok Imagine
- Photo editingNano Banana 2 Lite
- Clean downloads, no watermark
- Every model unlocked
- No shared queue
Secure checkout by Stripe. Card details never touch our servers.
Starter
What's included
- 3,150 credits per month
- Up to 630 images or 78 videosGrok ImagineNano Banana
- Clean downloads, no watermark
- Every model unlockedVeo 3.1SeedanceNano Banana
- Cost shown before you generate
- Generation history
Secure checkout by Stripe. Card details never touch our servers.
Creator
What's included
- 7,200 credits per month
- Up to 1,440 images or 180 videosGrok ImagineNano Banana
- Clean downloads, no watermark
- Every model unlockedVeo 3.1Seedance 2Nano Banana 2
- No shared queue
- Priority credit top-ups
Secure checkout by Stripe. Card details never touch our servers.
Studio
What's included
- 13,200 credits per month
- Up to 2,640 images or 330 videosGrok ImagineNano Banana
- Clean downloads, no watermark
- Every model unlockedSeedance 2.5Kling 3.0Veo 3.1
- No shared queue
- Built for 4K runs
Secure checkout by Stripe. Card details never touch our servers.
Credit packs
One-time top-ups — buy extra credits any time you run low.
Credit Pack 1500
What's included
- 1,500 credits that never expire
- Up to 300 images or 37 videosGrok ImagineNano Banana
- Clean downloads, no watermark
- No ongoing subscription
- Available across the supported models
- Credits never expire
Secure checkout by Stripe. Card details never touch our servers.
Credit Pack 3600
What's included
- 3,600 credits that never expire
- Up to 720 images or 90 videosGrok ImagineNano Banana
- Clean downloads, no watermark
- No ongoing subscription
- Available across the supported models
- Use for both video and image generation
- Credits never expire
Secure checkout by Stripe. Card details never touch our servers.
Credit Pack 9000
What's included
- 9,000 credits that never expire
- Up to 1,800 images or 225 videosGrok ImagineNano Banana
- Clean downloads, no watermark
- No ongoing subscription
- Handy for Seedance 2.0 final renders
- Credits never expire
Secure checkout by Stripe. Card details never touch our servers.
Charges appear as “KAVEL AI” on your card statement. You can cancel any time from Settings → Billing; cancellation takes effect at the end of the period you already paid for. Operator details, the full model list, and the refund window are on the about page.
Powered by
One credit balance for every Kavel image and video model
Use one balance across every image and video model on the shelf — the main ones are listed below — and check the credit cost before each request.
Video models
Image models
Model credit guide
What each model costs in credits
A failed run is refunded automatically. The exact estimate in the generator varies by model, length, resolution, audio, and number of images.
| Type | Model | Credit cost |
|---|---|---|
| Video | Seedance 2.5 | The long-take tier, priced per second: 5s 480p ≈ 525 credits, 5s 720p ≈ 1,185. A full 30s take runs ≈ 3,150 at 480p and ≈ 7,090 at 720p. |
| Video | Seedance 2.0 | Supplying a starting frame or clip costs less than starting from words alone: 6s 720p ≈ 565 credits from an image, ≈ 925 from text. Scales with resolution and length. |
| Video | Seedance 2 Fast | Faster and lower cost. 6s 720p text-to-video ≈ 745 credits. |
| Video | Seedance 2 Mini | The cheapest tier of the Seedance 2 family. 6s 720p ≈ 465 credits, 6s 480p ≈ 215 credits. |
| Video | Seedance 1.5 Pro | Audio doubles the rate. 6s 720p ≈ 85 credits silent, ≈ 160 with audio; 1080p ≈ 175 and ≈ 340. |
| Video | Veo 3.1 | Billed per video, not per second, and the tier you pick is the whole price: Lite ≈ 115 credits at 720p and ≈ 135 at 1080p, Fast ≈ 225 at 720p, Quality ≈ 940 at 720p. |
| Video | Kling 3.0 | Audio raises the rate by about half. 6s 720p ≈ 320 credits silent, ≈ 455 with audio; 1080p ≈ 405 and ≈ 610. |
| Video | MiniMax H3 | Fixed 2K, no resolution ladder. Priced per second — a 5s clip ≈ 395 credits. |
| Image | Nano Banana 2 | Generate or edit from text and images. 20 credits per 1K image, 30 at 2K, 45 at 4K. |
| Image | Nano Banana Pro | Consistent run times across generations. 30 credits per 1K or 2K image, 55 at 4K. |
| Image | Nano Banana 2 Lite | Faster, simpler variant, and the everyday editing price: a flat 15 credits per image at every resolution. |
| Image | GPT Image 2 | The lowest-cost premium image model here. 10 credits per 1K image, 15 at 2K, 30 at 4K. |
| Image | Seedream 5.0 Lite | Flat pricing — 20 credits per image whether you generate or edit. |
Questions about GPT Image 2.5
What shipped, what it costs, and what this page runs today.
GPT Image 2.5 launched on 8 September 2026 and no provider we buy inference from has listed it yet, so the generator on this page runs GPT Image 2 — free to start, no sign-up, and live right now. The day 2.5 appears on a provider, this page switches to it.
Flare is the fast tier — OpenAI's fastest model for high-quality everyday generation, at 50% lower latency than GPT Image 2. Sunburst is the precise tier, for premium workflows needing tighter control across edits, and it takes longer.
Both tiers share published rates: image output $30.00 per 1M tokens, image input $8.00 per 1M or $2.00 cached, text input $5.00 per 1M or $1.25 cached. Token counts per image are not published.
It runs GPT Image 2 and bills per image by size: roughly 10 credits at 1K, 15 at 2K and 30 at 4K. The estimate appears before you generate, and a failed run is refunded automatically unless it broke the content policy.
Not stated. OpenAI's launch material covers speed, editing behaviour and subject preservation, and publishes no resolutions or aspect ratios.
Yes — ChatGPT Images 2.5 is the consumer name, rolling out to ChatGPT, ChatGPT Work and Codex users across all tiers on desktop, mobile and web. GPT-Image-2.5 is the API side of the same release.
Three things ship with it: Sketch, which lets you draw inside ChatGPT and use the drawing as a reference; Templates for popular image formats; and comments placed on an image to aim an edit.
GPT Image 2 — it is the same family, the same instruction-led editing and three output sizes, and it is what the button on this page already runs. Start there; when 2.5 lands, your prompts carry straight over.
Generate now, free
Write a prompt or upload an image and go — free, no sign-up, with the credit estimate shown before you commit. It runs GPT Image 2 today, and GPT Image 2.5 the day a provider lists it.
Explore Kavel Models
Kavel guides & tutorials
Photo Scanning Services Cost in 2026: 9 Prices Compared
Nine photo scanning services priced per photo, read 9 Sep 2026. One has stopped taking orders entirely, and the cheapest and dearest per-photo figures come from the same company.
Best Free AI Image Generators 2026: Six Kinds of Free
Seventeen vendors' own pages, read 8 Sep 2026: seven no longer publish any number at all. Playground cut 1,000 a day to 10 per 3 hours. Plus who owns your free image.
Higgsfield AI Review 2026: What "Unlimited" Covers
Higgsfield's banner says unlimited; its help centre says specific models for a set period. Both are true, and the gap is where every complaint lives. Plans, user reports and rivals, read 9 Sep 2026.