Image sourcing
Control where new imagery comes from — nowhere, stock libraries, or AI generation.
Rivermark's design harness is based on HTML/CSS, which makes it token-efficient and gives it precise edits — but it needs an external image source for visual storytelling. The image_source parameter controls where the agent may pull new images from.
| Value | Behavior |
|---|---|
none | No new imagery. Only images present in the provided references can be incorporated into designs. |
stock | The agent pulls from a curated library of stock images. No attribution is required for stock usage. |
ai | Default. The agent knows when and how to generate images with AI image models, and how to place them into layouts. |
{
"task": "Create a LinkedIn post about our new warehouse.",
"image_source": "stock"
}Choosing a source
ai (default) — best visual quality and fit: the agent can generate imagery that matches the composition, palette, and aspect ratio it needs. Use it unless you have a reason not to.
stock — for products with strict "no AI imagery" policies, or where recognizable, real-world photography reads better. Stock usage requires no attribution.
none — for strict template fills and mechanical edits, where any new imagery would be wrong. Also the cheapest option, and the safest for brand-locked pipelines.
What's never approximated
Images in your references — product photos, logos, brand imagery — are used as-is at any image_source level. Rivermark may adapt them contextually where the design needs it (for example, recoloring a logo white so it reads on a dark background), but it never redraws or regenerates them with an AI model. image_source only governs imagery the agent would otherwise have to find or create.