Skip to main content

Image generation

Create a model with the image capability and an image_price (cost per image). An OpenAI-compatible image endpoint that returns b64_json works. For a dry run without a GPU, the offline demo seeds Offline demo images, which returns a synthetic placeholder.

  • Playground: switch to the Images tab, write a prompt, pick a size and a count.
  • API: POST /api/images {prompt, model?, n?, size?}, or the OpenAI-shaped POST /v1/images/generations.
  • Workflows: a generate_image step renders its prompt template (default {{input}}, and earlier step outputs are available) and makes one image.

Sizes are 256x256, 512x512, 1024x1024, 1024x1792 and 1792x1024, with up to four images per request. The prompt passes the guardrail policy, including the classifier model if one is set, and requests are refused once the day's budget is spent.

Images are stored as artifacts that expire after NUVORA_ARTIFACT_TTL_DAYS (default 7, up to 365). GET /api/artifacts/{id} serves one to the person who generated it and to administrators only. Every generation is metered and recorded as image.generated in the audit chain.