Test the workflow,
not the marketing claim.
Image jobs are expensive and slow enough that blind retries create duplicate cost and confusing assets, so the pattern distinguishes transport failures from rejected outputs and preserves an attempt history under one stable fixture ID.
Editorial product photograph of a matte ceramic skincare bottle on pale limestone, soft north-window light, delicate contact shadow, warm neutral palette, 85mm lens compression, no text, no logo.
Running shoe impact
Technical running shoe frozen above a red clay track at impact, fine dust particles suspended in air, directional hard sunlight, visible outsole detail, energetic diagonal composition, commercial sports photography.
Coastal house at dusk
Low modern coastal house at blue hour, weathered cedar and board-formed concrete, dune grasses moving in sea wind, warm interior light, realistic architectural photography, human-scale lens.
PYTHON QUICKSTART
Keep the request reproducible.
Store credentials in the environment. Record the exact model identifier, dimensions, test date, and retry count with the resulting asset.
import os
from openai import OpenAI
client = OpenAI(
api_key=os.environ["APIMART_API_KEY"],
base_url="https://api.apimart.ai/v1",
)
result = client.images.generate(
model="gpt-image-1-official",
prompt=PROMPT,
size="1024x1024",
n=1,
)Before you publish a result
- Record model ID, parameters, date, latency, and retry count.
- Review instruction following and protected attributes.
- Flag invented text, marks, anatomy, or unwanted objects.
- Keep the raw output linked to the original fixture.
- Do not claim a universal winner from one prompt.