How a design API works
Step 1
You send a request
A JSON payload with a text prompt or structured content, a brand kit reference, an output size, and a language. No visual input needed.
Step 2
The engine composes a design
A Large Design Model (LDM) selects layout patterns, places text as real editable text, positions images and vectors, and enforces brand rules: colors, fonts, logos, spacing. Every element is composed individually, not rendered as a flat image.
Step 3
You get back a layered design file
The response contains a structured design with every element on its own layer. Text is text. Vectors are vectors. You can move, edit, resize, or delete any element in the design studio. Export as JPG, PNG, HTML, or PDF.
Three types of design-related APIs
Not all APIs that produce visuals are the same. There are three distinct categories, each with a different architecture, output format, and use case.
Template APIs
How they work: You build a template with fixed slots (headline here, image there, logo in the corner). The API fills those slots with your data.
Output: A filled version of the template. Fixed layout, fixed structure.
Examples: Bannerbear, Placid, Abyssale
Good for: High-volume production of visuals that follow the exact same layout. Product cards, social media posts with identical framing, certificates.
Limitation: Every new layout requires a new template. Scaling to hundreds of design variations means building and maintaining hundreds of templates. No creative variety without manual template work.
Image generation APIs
How they work: A diffusion model generates a flat raster image from a text prompt. The model predicts pixels, not design elements.
Output: A flat JPG or PNG. No layers, no editable text, no structure.
Examples: DALL-E, Midjourney, Adobe Firefly, Stability AI
Good for: Generating artistic imagery, illustrations, concept art, background textures.
Limitation: Text in the image is rendered as pixels and frequently hallucinated (misspelled, garbled, or misplaced). No brand enforcement. No editability. Cannot produce marketing-ready designs where text accuracy and brand compliance matter.
Design APIs
How they work: A Large Design Model composes a design element by element. Text is placed as real text. Images and vectors are placed on separate layers. Brand rules (colors, fonts, logos) are enforced automatically.
Output: A multi-layered design file. Every element is individually addressable, editable, and exportable.
Examples: Sivi
Good for: Automated production of marketing designs — banners, social posts, ad creatives, email headers, product visuals — where text accuracy, brand compliance, and editability are required.
Advantage: No templates to maintain. Original layouts generated from scratch. Any size, any language (72+), full brand kit enforcement.
When to use a design API
A design API is the right choice when your use case requires any combination of the following:
Text accuracy
The design contains headlines, CTAs, product names, or prices that must be letter-perfect. No hallucinated text.
Brand compliance
Every output must use the correct hex codes, fonts, logo placement, and spacing. Enforced automatically, not manually.
Editability
The output needs to be editable after generation. Users or systems need to change text, swap images, or adjust layout.
Volume
You're generating hundreds or thousands of designs per day. Templates can't scale without proportional maintenance effort.
Multilingual
The same design needs to work in 10, 30, or 72 languages. The layout must re-compose per language, not just swap text strings.
Programmatic control
You need to generate designs from code, in a pipeline, or through an AI agent without any human in the loop.
Use cases by category
Any platform that needs to produce marketing visuals at volume, with text accuracy and brand compliance, is a candidate for a design API. Common categories include:
How Sivi's design API works
Sivi is the first design API built on a Large Design Model. Here's what that means for developers:
Input options
Designs from Prompt: Send a natural language prompt ("Summer sale banner for urban clothing brand") and Sivi composes the design.
Designs from Content: Send structured content (headline, body text, CTA, product image URL) and Sivi arranges it into a design.
Any size, any language
Specify any custom dimension. The model doesn't stretch or crop, it composes the entire layout for the target size. Same for languages: switch from English to Arabic, and the layout flips to RTL with correct typography.
Brand kit enforcement
Attach a brand kit to your API key. Every design generated through that key automatically uses the correct colors, fonts, logos, and spacing. No manual checking.
Three integration paths
When a design API is NOT the right choice
You need artistic imagery.
For photo-realistic scenes, illustrations, concept art, use an image generation API like DALL-E or Midjourney.
You need photo editing.
For cropping, retouching, filters, background removal, use an image editing API, not a design API.
You need a fixed, pixel-perfect layout every time.
If the design must look identical across every output with zero variation, a template API gives you that control.
Design APIs: What everyone wants to know
Is a design API the same as an image generation API?
No. An image generation API (DALL-E, Midjourney, Firefly) produces a flat raster image where text is pixels and nothing is editable. A design API produces a structured, multi-layered design where text is real text, every element is on its own layer, and the entire output is editable. Different architecture, different output, different use case.
Can a design API replace a graphic designer?
For high-volume, templated work - yes. For bulk product banners, social post variations, ad creatives in multiple sizes and languages, a design API handles this faster and cheaper than manual design. For strategic brand work, campaign concepts, and custom creative direction, you still need a designer.
Do I need templates to use a design API?
No. A generative design API like Sivi creates original layouts from scratch. You send a prompt or structured content, and the model composes a new layout every time. No templates to build or maintain.
How does a design API handle brand consistency?
Through a brand kit. You upload your brand's colors, fonts, logos, and style rules once. The API enforces them on every design generated under that brand kit. No manual review needed.
Can a design API generate designs in multiple languages?
Sivi supports 72+ languages including RTL scripts (Arabic, Hebrew, etc.). It composes the entire layout so typography, reading direction, and spacing work correctly in the target language.
How do AI agents use a design API?
Through the Model Context Protocol (MCP). The AI agent treats the design API as a callable tool. It sends a design brief, receives a layered design back, and presents it to the user. This means any MCP-compatible assistant (Claude, ChatGPT, Cursor, Windsurf) can generate marketing designs as part of a conversation without switching apps.
What's the cost of using a design API?
Sivi's design API starts with a $1 sandbox (20 credits, 7-day validity). Plans start at $96/month for 800 designs when paid yearly. At scale, that's roughly $0.11 per design, compared to $15+ per banner with manual design at $60/hour. View full pricing
