gpt-image-2 API

gpt-image-2 sits at the top of OpenAI's image model family, above gpt-image-1.5 and gpt-image-1-mini. Its standout traits are strong prompt adherence on complex scenes, photorealistic lighting and materials, and — the thing that made the gpt-image line famous — reliably legible text rendered inside the image, which most diffusion-style models still get wrong. Per OpenAI's model spec it accepts text and image inputs and returns images, so it covers both fresh generation and guided editing workloads. On CaMeL Hub the model is billed per call rather than per token: every successful generation costs a flat $0.20, deducted from your balance. That makes budgeting trivial for batch pipelines — 500 marketing assets cost exactly $100, with no surprises from resolution or quality settings inflating output-token counts. You top up by Alipay, WeChat Pay, or international cards, and the same API key that runs your chat models runs this one. Behavior-wise, gpt-image-2 is not a chat model. You reach it through the dedicated images endpoint (client.images.generate with model="gpt-image-2"), it does not stream, and it does not do function calling. Responses come back as image data (base64), ready to write straight to disk or pipe into the next stage of an agent workflow.

curl https://api.camel-hub.com/v1/images/generations \
  -H "Authorization: Bearer $CAMEL_HUB_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2",
    "prompt": "A red panda coding at night, cinematic lighting",
    "n": 1,
    "size": "1024x1024"
  }'

from openai import OpenAI

client = OpenAI(
    api_key="$CAMEL_HUB_KEY",
    base_url="https://api.camel-hub.com/v1",  # <-- the only change vs. the OpenAI default
)

img = client.images.generate(
    model="gpt-image-2",
    prompt="A red panda coding at night, cinematic lighting",
    size="1024x1024",
    n=1,
)
print(img.data[0].url)