Image generation and edits

Image models use the OpenAI Images API. The request waits until the images are ready and returns them in one response. You pay per image returned; images that fail are not charged.

Find model IDs and prices under Models → Image. Availability depends on your key's group: GET /v1/models lists what your key can call.

#Generate an image

curl https://bigmodel.io/v1/images/generations \
  -H "Authorization: Bearer $BIGMODEL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen-image-2.1",
    "prompt": "A paper lantern floating over a misty lake at dawn, the words \"Good Morning\" painted on it",
    "size": "1536x1024",
    "n": 1
  }'

import os
from openai import OpenAI

client = OpenAI(base_url="https://bigmodel.io/v1", api_key=os.environ["BIGMODEL_API_KEY"])

result = client.images.generate(
    model="qwen-image-2.1",
    prompt="A paper lantern floating over a misty lake at dawn",
    size="1536x1024",
)
print(result.data[0].url)

import OpenAI from "openai";

const client = new OpenAI({ baseURL: "https://bigmodel.io/v1", apiKey: process.env.BIGMODEL_API_KEY });

const result = await client.images.generate({
  model: "qwen-image-2.1",
  prompt: "A paper lantern floating over a misty lake at dawn",
  size: "1536x1024",
});
console.log(result.data[0].url);

The response is {"created": 1760000000, "data": [{"url": "https://bigmodel.io/v1/media/..."}]}. The link works without an API key for 24 hours; download the file if you need it longer, and treat the link as private.

#Request fields

stream is not supported for these models.

#Edit an image

Send one or more input images with an instruction. Inputs can be public https:// URLs or base64 data URIs (data:image/png;base64,...), up to 50 MB each and 10 per request.

curl https://bigmodel.io/v1/images/edits \
  -H "Authorization: Bearer $BIGMODEL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen-image-2.1",
    "prompt": "Make it a snowy winter evening, keep the composition",
    "images": [{"image_url": "https://example.com/street.jpg"}]
  }'

import os
from openai import OpenAI

client = OpenAI(base_url="https://bigmodel.io/v1", api_key=os.environ["BIGMODEL_API_KEY"])

with open("street.jpg", "rb") as image:
    result = client.images.edit(
        model="qwen-image-2.1",
        image=image,
        prompt="Make it a snowy winter evening, keep the composition",
    )
print(result.data[0].url)

Multipart uploads (image, or image[] for several files) work too, which is what the OpenAI SDKs send. With several inputs, the first is the image to edit and the others are references.

Tool models take an input image and no prompt: image-upscale returns a larger copy, image-remove-background returns the subject on a transparent background.

#Use an image as a video's first frame

Most video models animate a first frame. Generate it here, then pass its URL as image to POST /v1/videos.

#Verify the setup

Run the generation example. A 200 with a data[0].url that opens an image means your key can use image models. The charge appears in Usage as one line per request.

#Troubleshooting

404: the model is not available. The model is not offered in your key's group. Check GET /v1/models, or use a key in a group that includes image models.

400 invalid_request_error: "An input image could not be read". The URL is not public https://, or the data URI is malformed or larger than 50 MB.

400: "This model edits images: send an input image". The model only edits. Call /v1/images/edits with an input image.

400: "n must be at most 4". Ask for at most 4 images per request.

502 upstream_error. The image could not be generated, often because of the prompt or input. The message says why when it is known. You are not charged for failed images.

403 billing_error. Your balance is used up. Top up in Billing.

#Related

• Video generation to animate the image.
• Image input (vision) to send images to chat models instead.
• Pricing for every image price.