Pictures

Make a picture

POST /v1/images/generations

from openai import OpenAI

client = OpenAI(base_url="http://localhost:7860/v1", api_key="not-needed")
result = client.images.generate(prompt="a lighthouse at dawn, watercolour", size="1024x1024")
print(result.data[0].url)

The request waits until the picture is finished. By default you get a URL on wanna.sh (http://localhost:7860/files/...); ask for "response_format": "b64_json" to get the image inline.

Field	Notes
prompt	Required.
model	auto, a model id or a Hugging Face repo id.
n	How many pictures. Default 1.
size	"WIDTHxHEIGHT", for example "1024x768", or "auto" for the model’s usual size.
quality	low, medium, high (OpenAI’s names), or wanna.sh’s fast, balanced, best.
response_format	url (default) or b64_json.

wanna.sh also accepts these extra fields (with the OpenAI SDKs, pass them in extra_body):

Field	Notes
negative_prompt	What to avoid.
steps (or num_inference_steps)	Number of steps. Overrides quality.
guidance_scale	How closely to follow the prompt.
seed	Same seed and prompt give the same picture.
sampler	dpmpp_2m_karras, euler_a, euler, unipc, ddim, or builtin for the model’s own. Works with Stable Diffusion style models; a model that can’t use the sampler you name (FLUX, for example) returns an error.

result = client.images.generate(
    prompt="a lighthouse at dawn, watercolour",
    extra_body={"negative_prompt": "text, watermark", "seed": 42, "steps": 30},
)

Restyle a picture

POST /v1/images/edits

Multipart form upload. Starts from your picture and changes it to match the prompt.

curl http://localhost:7860/v1/images/edits \
  -F image=@photo.png \
  -F prompt="make it look like a pencil sketch" \
  -F strength=0.6

Field	Notes
image	Required. The starting picture.
prompt	Required.
model, n, response_format, seed	As above.
strength	0 to 1. How far it may move from the original. Default 0.6.

Masks (painting over only part of a picture) aren’t supported yet.