Image

POST /v1/images/generations

Synchronous — the server polls MuleRun internally until the image is ready, then returns OpenAI's standard { data: [{ url }] } envelope.

curl http://localhost:51222/v1/images/generations \
  -H "Authorization: Bearer $CLI2API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"wan2.6-t2i","prompt":"a synthwave fox","size":"1024x1024","n":1}'

Supported model: gpt-image-2, nano-banana, nano-banana-pro, nano-banana-2, midjourney, wan2.6-t2i, wan2.6-image, wan2.5-t2i-preview, wan2.5-i2i-preview.

gpt-image-2 image quality

gpt-image-2 generation accepts a quality parameter with low, medium, high, and auto as valid values. MuleRun defaults to high when the field is omitted. Output dimensions are controlled separately with size.

curl http://localhost:51222/v1/images/generations \
  -H "Authorization: Bearer $CLI2API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-image-2","prompt":"a vector fox logo","quality":"low","size":"1024x1024","n":1}'

from openai import OpenAI
c = OpenAI(api_key="local-key", base_url="http://localhost:51222/v1")

r = c.images.generate(
    model="gpt-image-2",
    prompt="a vector fox logo",
    quality="low",
    size="1024x1024",
)
print(r.data[0].url)

POST /v1/images/edits

Edit an existing image with a text prompt. Accepts both application/json (images as URLs / data: URIs) and multipart/form-data (OpenAI's native file uploads).

curl http://localhost:51222/v1/images/edits \
  -H "Authorization: Bearer $CLI2API_KEY" -H "Content-Type: application/json" \
  -d '{"model":"nano-banana-edit","prompt":"add sunglasses","images":["https://input/portrait.png"]}'

curl http://localhost:51222/v1/images/edits \
  -F "model=gpt-image-2-edit" -F "image=@photo.png" \
  -F "prompt=oil painting" -F "size=1024x1024"

Supported model (always -edit suffix to disambiguate from generation): gpt-image-2-edit (supports mask), nano-banana-edit, nano-banana-pro-edit, nano-banana-2-edit, wan2.5-i2i-preview-edit.