Overview
Programmatically create and manipulate on-brand designs at scale.
Rivermark's Design API puts a frontier-class design agent behind a single HTTP call. You describe the task in natural language, provide references — brand kits, images, webpages, HTML, zip archives, existing designs — and get back finished, rendered designs.
Behind that one call, Rivermark runs agent sessions, reads your references, renders its work, reviews the output, and corrects it before it reaches you. You get finished designs without building or running any of that infrastructure yourself.
What you can build
AI design editors
A textbox and an attachment box is all you need. Turn a user's words and files into a single design they can keep editing conversationally.
On-brand content automation
Extract a user's brand once, then generate unlimited on-brand posts, ads, and assets from copy or a content plan.
User-created templates
Create a template once, then fill it with new copy for every post at a fraction of the cost.
Multi-format resizing
Take any design and adapt it to every format and platform, with intelligent reflow rather than naive cropping.
Why it's different
- One interface for every task. There is no template schema to learn and no fixed capability matrix. Creating, editing, resizing, remixing, and filling in copy are all just instructions. When your requirements change, rewrite the prompt.
- Built on HTML and CSS. Every design is a real HTML/CSS document. This provides exceptional consistency across asset sets and output that is deterministic, portable, and editable.
- Exact brand adherence. Your — or your users' — fonts, hex values, logos, and imagery are used as-is, never approximated or redrawn. Brand kits extracted from websites, guidelines, and Canva designs feed directly into every request.
- Persistent, addressable designs. Every request returns a
design_idyou can reference in later requests. Designs are first-class, long-lived objects: use them as templates, fork them into editing sessions, or let your users iterate on them. - Costs that scale with the work. You pay for what the agent does. A mechanical template fill bills at a fraction of a new design from scratch.
How it works
- You send a task, optional references, and options like creativity level and image sourcing.
- The agent works — reading references, browsing allowed webpages, writing and editing HTML/CSS, generating or pulling imagery, and rendering the design.
- It reviews itself. A review pass catches poor legibility, overflow, clipping, invented content, and edit failures, and fixes them before you see output.
- You get rendered frames, a persistent design ID, a summary of what changed, and the exact cost of the call.
Progress streams over Server-Sent Events, so you can build engaging loading states with real textual progress and draft previews.
Ready to start?
The getting started guide takes you from an API key to your first rendered design in a few minutes, with switchable instructions for TypeScript, Python, and cURL.