An approved image shouldn't turn into a guessing game on the next run. Keep the prompt, dimensions, tier, and seed with the job record, then send the same values when the workflow needs the asset again. 

{
  "prompt": "Studio product photo of a matte black smart notebook on a warm off-white surface, soft side light, no text",
  "width": 1024,
  "height": 1024,
  "tier": "fast",
  "seed": 481516
}

POST that body to /image-generate. The response returns the hosted image_url along with the resolved model, final dimensions, prompt, and seed. Save that response beside the asset. The next production step can reuse the same request or change one field deliberately, such as the prompt or size.

Use fast for routine image work. Pick creative for a more stylized render, anime for illustration, or sd35 when a Stable Diffusion-style output is the right fit. Keep the tier fixed when you need a comparable rerun, since it chooses the model behind the request.

The route accepts negative_prompt when a batch needs a shared exclusion, and it snaps dimensions to the supported 256 to 1024 pixel range in 16-pixel increments. That keeps stored jobs valid without a separate image setup step.

This is useful for catalog images, recurring social creative, and any agent flow that needs to regenerate a known visual without a vendor account or file upload.