GPT Image 2.5 API : Generate AI Images up to 4K, With Transparent Backgrounds
Try GPT Image 2.5 online right here: write a prompt, add up to 16 reference images, choose the quality, resolution and aspect ratio, and get up to 4 images at once. Then call the same model from your product through the GPT Image 2.5 API, with one key and one balance.
- Up to 16 reference images
- Five quality levels
- Billed per image
- Open GPT Image 2.5 API
Features
Key GPT Image 2.5 API Capabilities
Text to image, guided by references
Describe the picture in a prompt of up to 8,000 characters, or add up to 16 reference images and name each one as @Image1, @Image2 and so on. Without image_urls it's pure text to image; with them you can, for example, place the product from @Image1 on a marble table.
Five quality levels
quality goes from low through medium, high and xhigh to max. Together with the resolution it sets the price per image, so you can draft at low and render the final version at the level the job needs.
Transparent backgrounds
Set background to transparent and GPT Image 2.5 returns a PNG with a transparent background, ready for logos, stickers, icons and product cut-outs, with or without reference images. It has to be the parameter: asked for only in the prompt, the model paints a checkerboard into an opaque image. It doesn't change the price.
Two variants: Flare and Sunburst
Pick variant flare or sunburst for each task, at the same price. Flare is fast, great for everyday creation and high-volume generation; Sunburst gives more precise details and edits for polished, final-quality images, and takes longer.
1k to 4k, 15 aspect ratios
Render at 1k, 2k or 4k in 1:1, 3:2, 2:3, 4:3, 3:4, 4:5, 5:4, 16:9, 9:16, 21:9, 27:16, 16:27, 9:8 or 8:9, or leave aspect_ratio on auto to follow your reference images (1:1 without them).
Up to 4 images per task
Set num_outputs from 1 to 4. The images are generated in parallel, so a set of four takes about as long as one, and at low quality and 1k a single image takes about 15 seconds.
Overview
What Is GPT Image 2.5?
- Model ID
gpt-image-2.5- Resolution
- 1k / 2k / 4k
- Quality
- Low / Medium / High / X-High / Max
- Aspect ratio
- 15 options
- Reference images
- Up to 16
- Images per task
- 1–4
- Prompt
- Up to 8000 characters
- Price
- From $0.0117 / image
GPT Image 2.5 is an AI image model that generates pictures from a prompt, optionally guided by reference images, with a quality level you choose and an optional transparent background. Prompts can run up to 8,000 characters and name each reference as @Image1, @Image2 and so on, so one request can say exactly which product, character or style goes where.
On NanoAPI, GPT Image 2.5 is the model gpt-image-2.5 of the product gpt_image, in two variants, Flare and Sunburst, at the same price. Every task takes a prompt and up to 16 reference images and returns 1 to 4 images, each as a full PNG plus a JPEG preview of up to 1024 pixels on the long side, on public links that don't expire. You choose the variant, quality, resolution, aspect ratio and number of images for every task.
Quality is the setting that matters most: low gives a quick image for drafts, previews and high-volume jobs, while high, xhigh and max put more into each picture for the final asset. Quality and resolution together set the price per image, so you choose the trade-off task by task instead of committing to one plan.
Use GPT Image 2.5 online in the studio at the top of this page, then call the GPT Image 2.5 API at https://api.nanoapi.ai/v1 from your own app. Both draw on the same prepaid balance at the same price per image, and you only pay for the images you get: an image that isn't generated is refunded, and so is a task that fails.
Pricing
GPT Image 2.5 API Pricing
Priced per image by quality and resolution: you're charged only for the images a task returns, and variant and background don't change the price.
| Quality | 1k / image | 2k / image | 4k / image |
|---|---|---|---|
Lowlow | $0.0117 | $0.0234 | $0.0351 |
Mediummedium | $0.0234 | $0.0468 | $0.0585 |
Highhigh | $0.0702 | $0.1287 | $0.1989 |
X-Highxhigh | $0.117 | $0.2106 | $0.3276 |
Maxmax | $0.234 | $0.4212 | $0.702 |
Live prices per image. Cost = price per image × number of images; you're only charged for the images you get, and failed runs are fully refunded.
How it works
How to Use the GPT Image 2.5 API
- 1
Step 1: Get your API key
Sign in to NanoAPI with Google and create a key on the API Keys page of the console. It authenticates every GPT Image 2.5 API request, so keep it on your server.
- 2
Step 2: Top up your balance
Add funds on the Top up page of the console. The balance is prepaid in USD, never expires and is shared by every NanoAPI model, so one balance covers images and videos alike.
- 3
Step 3: Create a task
Send POST /v1/tasks with model gpt-image-2.5, a prompt, variant, quality, resolution, aspect_ratio and num_outputs, plus image_urls and background if you need them. The task is queued in under a second, and the price of the requested images is reserved within a few seconds.
- 4
Step 4: Get the images
Receive the finished task once at your webhook_url, or poll GET /v1/tasks/{task_id} every 15 seconds or more. At low quality and 1k an image takes about 15 seconds; the PNG files are in outputs[].url and the JPEG previews in preview_url.
Why NanoAPI
Why Use the GPT Image 2.5 API on NanoAPI
One API for images and video
GPT Image 2.5 uses the same POST /v1/tasks endpoint, task object, API key and prepaid balance as every NanoAPI model, so adding another image or video model never means a new integration.
Unfinished images aren't billed
A task reserves the price of the images you ask for and charges only for the ones it returns. If moderation blocks one of four, the task still succeeds and that image is refunded; if none come back, the whole amount is.
Webhooks, signed
Pass a webhook_url and receive the finished task once, signed to the Standard Webhooks spec so you can verify it if you want to. Or poll one task, or a batch of task ids in one request.
No double charges
Send an Idempotency-Key when you create a task: if a request times out and you resend it, you get the task that was already created instead of a second one.
Spend limits per key
Cap each API key's hourly, daily or total spending in the console, so a bug or a busy day can't use up your balance. Each request is listed with its status, inputs, cost and result.
Try online, then integrate
The studio on this page runs the same GPT Image 2.5 model at the same prices, so you can settle on a quality and resolution before you write code. The docs also come as one llms.txt file for AI coding assistants.
Use cases
What You Can Build With the GPT Image 2.5 API
Logos, icons and stickers
Generate assets with background set to transparent and drop the PNG straight into a design, an app or a print file, with no cut-out step.
Product photos and ads
Send the product as @Image1 and place it on a marble table, in a kitchen or on a poster, with up to 16 references to hold the brand's look.
Fast drafts, polished finals
Explore ideas at low quality, where a 1k image takes about 15 seconds, then render the chosen prompt at high, xhigh or max for the final asset.
Image generation in your product
Add GPT Image 2.5 to a design tool, a marketplace or a content pipeline: create tasks in bulk, get the images by webhook and serve the PNG links as they are.
FAQ
GPT Image 2.5 API FAQ
Can’t find what you need? Write to us.
01What is the GPT Image 2.5 API?
The GPT Image 2.5 API lets your app generate images with the GPT Image 2.5 model over HTTP. On NanoAPI you send POST /v1/tasks with model gpt-image-2.5, a prompt and the image settings, optionally with up to 16 reference images, and get 1 to 4 images back by webhook or by polling.
02How much does the GPT Image 2.5 API cost?
You pay per image: the price of the quality and resolution you choose × the number of images generated. A task reserves the price of every image in num_outputs, charges only for the images it returns and refunds the rest; a failed task is refunded in full. variant and background don't change the price. The live prices, per quality and resolution, are in the table above, on the pricing page and in GET /v1/products.
03How do I get a transparent background?
Set background to transparent; it works with reference images too. The PNG at outputs[].url keeps the transparency, while the JPEG preview at preview_url is placed on white. Leaving background out, auto and opaque all give an opaque image, and writing “transparent background” in the prompt alone only paints a checkerboard into it. None of this changes the price. Reference images with transparency are placed on white before they're used.
04What are the Flare and Sunburst variants?
Two versions of GPT Image 2.5 that you choose with variant, with the same parameters and price. Flare is fast, great for everyday creation and high-volume generation. Sunburst gives more precise details and edits for polished, final-quality images, and takes longer.
05How many reference images can I use, and in which formats?
Up to 16 per task, as public http(s) links in image_urls, named in the prompt as @Image1, @Image2 and so on in that order. JPEG, PNG, WebP, AVIF, GIF (the first frame), BMP and TIFF work. HEIC and HEIF don't, so convert iPhone photos to JPEG first; the studio on this page does it for you.
06Which resolutions and aspect ratios does GPT Image 2.5 support?
resolution is 1k, 2k or 4k. aspect_ratio is auto, 1:1, 3:2, 2:3, 4:3, 3:4, 4:5, 5:4, 16:9, 9:16, 21:9, 27:16, 16:27, 9:8 or 8:9. auto follows your first reference image, or gives 1:1 when there are none. quality and variant don't change the size.
07What happens if only some of my images are generated?
The task still succeeds and outputs contains only the images that were made, so it can be shorter than num_outputs; this happens, for example, when moderation blocks one image. You're charged for the images you got and the rest goes back to your balance. Only when no image comes back does the task fail, with a full refund.
08How long does GPT Image 2.5 take, and can I try it first?
At low quality and 1k an image takes about 15 seconds, and several images run in parallel, so a task of four takes about as long as one. To try it, sign in with Google and use the studio at the top of this page: same model, same prices and the same balance as the GPT Image 2.5 API.
Start Creating With GPT Image 2.5
Try GPT Image 2.5 in the studio above, then use the GPT Image 2.5 API to bring the same model into your product.