GPT Image 2 - AnyFast

GPT Image 2 is OpenAI’s image generation and editing model. AnyFast exposes the public model ID gpt-image-2 through two Image API endpoints:

POST /v1/images/generations creates images from text.
POST /v1/images/edits edits one or more uploaded images, with an optional mask.

Select the Direct resource group when using gpt-image-2. GPT Image models return Base64-encoded image data in data[].b64_json; do not send the DALL·E-only response_format parameter.

Key capabilities

Capability	How to use it	Important constraint
Text-to-image	Send a JSON request to `/v1/images/generations`	Prompt length: `1`–`32,000` characters
Transparent output	Set `background` to `transparent`	Preview feature; use PNG or WebP output
Image editing	Upload `image` or as many as 16 `image[]` files to `/v1/images/edits`	Use multipart form data
Mask editing	Add an alpha-channel `mask` to an edit request	The mask applies to the first source image
Flexible dimensions	Set `size` to `auto` or `WIDTHxHEIGHT`	Custom dimensions must satisfy the size rules below

Capability guide

The images below come from OpenAI’s shared image generation guide, which currently illustrates GPT Image 2.5. They show generation and editing workflows; requests on this page still use gpt-image-2.

Generate images
Transparent output
Edit images
Mask edits

Describe the desired subject, composition, lighting, style, and text in prompt. The model supports low, medium, high, and auto quality settings and can return from one through ten images.Square images are typically fastest. Use a custom size when the final layout matters, but keep both edges divisible by 16.

A beige coffee mug on a wooden table in OpenAI's image generation guide — Coffee mug — OpenAI guide example

A river made of white owl feathers in a winter landscape — Feather river — OpenAI guide final image

The feather-river prompt combines a subject, material, and setting: a river made of white owl feathers, winding through a winter landscape. Only the final image is shown here.

Set background to transparent and request png or webp. Include the transparency requirement in the prompt as well, for example, “isolated icon on a fully transparent canvas with no shadow.”

OpenAI marks transparent-background support for GPT Image 2 as a preview feature. Use PNG or WebP output when requesting a transparent background.

Upload one source image with image, or multiple sources with repeated image[] fields. GPT Image 2 accepts as many as 16 source images and processes image inputs at high fidelity automatically; omit input_fidelity.Use the prompt to say what must change and what must remain unchanged. The edit endpoint returns a newly generated image rather than modifying the source file in place.

A gray cat holding an otter wearing an orange scarf — Before — OpenAI's source image

The cat and otter transformed into a photorealistic image — After — OpenAI's realistic-style edit

OpenAI’s multi-turn example requests a more realistic style in a follow-up prompt. This comparison illustrates the visual edit; with the Image API documented here, upload the source image to /v1/images/edits and describe the desired change.

Add mask to guide a localized edit. The mask must contain an alpha channel, match the first source image’s dimensions and format, and be smaller than 50 MB. Transparent mask regions identify the area to edit.A mask guides the model but is not a guaranteed pixel-exact boundary. With multiple source images, it applies only to the first image.

A sunlit lounge with an indoor swimming pool before editing — Source image

The lounge mask with a transparent area over part of the pool — Mask — area to edit

The lounge pool after adding an inflatable flamingo in the masked area — OpenAI's edited output

OpenAI’s mask-edit example adds an inflatable flamingo to the pool. The source provides the scene, the transparent mask region identifies the area to change, and the prompt describes the desired result.

Request examples

Generate
Transparent PNG
Edit
Mask edit

cURL

curl --request POST \
  --url https://www.anyfast.ai/v1/images/generations \
  --header 'Authorization: Bearer YOUR_API_KEY' \
  --header 'Content-Type: application/json' \
  --data '{
    "model": "gpt-image-2",
    "prompt": "A clean editorial product photograph of a cobalt glass bottle on pale stone, soft morning light",
    "size": "1536x1024",
    "quality": "high",
    "output_format": "jpeg",
    "output_compression": 85,
    "n": 1
  }'

cURL

curl --request POST \
  --url https://www.anyfast.ai/v1/images/generations \
  --header 'Authorization: Bearer YOUR_API_KEY' \
  --header 'Content-Type: application/json' \
  --data '{
    "model": "gpt-image-2",
    "prompt": "A simple green circular app icon, isolated on a fully transparent canvas, no shadow",
    "size": "1024x1024",
    "quality": "low",
    "background": "transparent",
    "output_format": "png",
    "n": 1
  }'

cURL

curl --request POST \
  --url https://www.anyfast.ai/v1/images/edits \
  --header 'Authorization: Bearer YOUR_API_KEY' \
  --form 'model=gpt-image-2' \
  --form 'image[]=@/path/to/product.png' \
  --form 'image[]=@/path/to/reference.png' \
  --form 'prompt=Keep the product unchanged and apply the lighting and color palette from the reference image' \
  --form 'size=1536x1024' \
  --form 'output_format=png' \
  --form 'n=1'

cURL

curl --request POST \
  --url https://www.anyfast.ai/v1/images/edits \
  --header 'Authorization: Bearer YOUR_API_KEY' \
  --form 'model=gpt-image-2' \
  --form 'image=@/path/to/source.png' \
  --form 'mask=@/path/to/mask.png' \
  --form 'prompt=Replace only the masked background with a warm studio gradient' \
  --form 'size=1536x1024' \
  --form 'output_format=png' \
  --form 'n=1'

Output controls

Parameter	Values	Notes
`quality`	`low`, `medium`, `high`, `auto`	Defaults to `auto`; higher quality generally uses more image output tokens
`size`	`auto` or `WIDTHxHEIGHT`	Defaults to `auto`; common sizes include `1024x1024`, `1536x1024`, and `1024x1536`
`background`	`auto`, `opaque`, `transparent`	Transparent output is in preview for GPT Image 2 and requires PNG or WebP
`output_format`	`png`, `jpeg`, `webp`	Defaults to PNG
`output_compression`	`0`–`100`	Applies only to JPEG and WebP; defaults to `100`
`moderation`	`auto`, `low`	Defaults to `auto`
`n`	`1`–`10`	Defaults to `1`

Custom size rules

Custom WIDTHxHEIGHT dimensions must satisfy all of these rules:

Width and height are multiples of 16 pixels.
The aspect ratio is between 1:3 and 3:1.
Neither edge exceeds 3,840 pixels.
Total pixels are between 655,360 and 8,294,400.
Resolutions above 2560x1440 are experimental.

Image editing inputs

Upload one image with image, or as many as 16 images with image[]; do not send both forms in one request.
Use PNG or JPEG source files through multipart form uploads.
Each source image and optional mask must be smaller than 50 MB.
GPT Image 2 always processes edit inputs at high fidelity, so the request must omit input_fidelity.

Response handling

The AnyFast response contains created, model, data, and token usage. Decode data[0].b64_json and save it with the extension selected by output_format.

Decode the first result

echo "$RESPONSE" | jq -r '.data[0].b64_json' | base64 --decode > result.png

For transparent output, save the Base64 data with a .png or .webp extension that matches output_format.

Official capability gallery

The workflow images above are embedded from OpenAI’s API guide. For GPT Image 2-specific examples of text-heavy layouts, multilingual text, flexible aspect ratios, and controlled edits, see the launch gallery below.

Explore the GPT Image 2.0 launch gallery

View OpenAI’s official GPT Image 2.0 generation and editing showcase.

Limitations

Complex prompts can take up to two minutes to process.
Precise text placement and clarity can still require iteration.
Recurring characters and brand elements can vary across separate generations.
The model may struggle with exact composition control in highly structured layouts.

Looking for a lower-cost AnyFast variant with response_format support? See gpt-image-2-c. It does not support the n parameter.

See the OpenAI GPT Image 2 model page, image generation guide, create image reference, and create image edit reference for the upstream specification.

GPT Image 2 API Reference

View generation and editing request fields and response schemas.

​Key capabilities

​Capability guide

​Request examples

​Output controls

​Custom size rules

​Image editing inputs

​Response handling

​Official capability gallery

Explore the GPT Image 2.0 launch gallery

​Limitations

GPT Image 2 API Reference

Key capabilities

Capability guide

Request examples

Output controls

Custom size rules

Image editing inputs

Response handling

Official capability gallery

Limitations