AI agents are getting powerful at text - but they still can't make images. No social posts, no banners, no infographics. I built RendrKit to fix that.
What It Does
One API call with a text prompt returns a production-ready PNG:
bash
curl -X POST https://api.rendrkit.dev/api/v1/generate \
-H "Authorization: Bearer YOUR_KEY" \
-H "Content-Type: application/json" \
-d '{"prompt": "Launch announcement for RendrKit - Design API for AI agents"}'
Behind the scenes:
1. GPT-4o-mini analyzes the prompt and picks the best template
2. It fills in the template slots (headline, subtitle, colors, photo query)
3. Playwright renders the HTML/CSS to PNG
4. Image goes to CDN, you get a URL back
Two Modes
Prompt Mode - send text, get image. GPT does the thinking. $0.005/image.
Direct Render - you pick the templateId and fill slots yourself. No GPT, no latency. $0.001/image.
The Stack
• Next.js on Vercel for API routes
• Playwright for HTML-to-PNG rendering
• GPT-4o-mini for template selection + slot filling
• Supabase for auth, usage tracking, API keys
• Unsplash/Pexels for background photos
50+ Templates
Social posts, banners, cards, infographics, event posters, testimonials, real estate listings - all responsive across 5 sizes (1080x1080, 1200x628, 1080x1920, 1200x1200, 1280x720).
Integrations
• REST API
• npm SDK (@rendrkit/sdk)
• MCP server (@rendrkit/mcp)
• OpenAI function calling schema
What I Learned
• Playwright on serverless is tricky but doable (Vercel has a Chromium layer)
• GPT-4o-mini is good enough for template selection at $0.005/image margin
• HTML/CSS templates beat AI image generation for structured content - predictable, fast, brandable
Check it out: rendrkit.dev
Free tier: 50 images/month. Would love feedback!