Title to Thumbnail Generation

Send a title and a format. Get back a production-ready thumbnail in WebP or PNG in under 25 seconds. No templates to design. No variables to bind. No layout file to maintain. Every generation is a fresh image composed by an AI pipeline that reads your title and produces something on-brief.

POST /v1/generate

{
  "title": "10 crypto scams",
  "format": "youtube"
}

AI Pipeline

A real POST /v1/generate call with a short YouTube title, and the actual AI-generated thumbnail returned in the response.

Templates vs. AI Generation

Most "thumbnail APIs" are actually template-fill APIs. You design a layout in a visual editor once (title placeholder, image slot, background), then hit an endpoint that fills that layout with per-request variables. The layout never changes. Every output looks like the template with different text.

ThumbAPI works differently. There is no layout you build. There is no template you maintain. Every generation is a fresh image composed by an AI pipeline. Two different titles in the same category produce two different-looking thumbnails in the same visual family with a different composition.

When templates win

• You need pixel-identical output across thousands of images (e.g. product cards where every card must be the same layout)
• You have a designer already producing templates and want to industrialize their output
• You need SVG or PDF export for print or vector use

When AI generation wins

• You want visual variety — every YouTube video should feel unique, not stamped from the same template
• You do not have a designer and do not want to become one
• You publish across multiple niches where a single template cannot possibly fit all content
• You want thumbnails that track current trends without rebuilding your template library every quarter

The Simplest Possible Call

Send a title and a target format. That is it. Every other parameter has a sensible default.

curl -X POST https://api.thumbapi.dev/v1/generate \
  -H "x-api-key: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "title": "How to grow a YouTube channel from 0 to 10K",
    "format": "youtube"
  }'

How the Pipeline Works

Three stages happen inside a single request:

1. Classification — an LLM reads your title, picks the most relevant niche category (or uses the one you passed), and extracts salient signals (numbers, entities, emotional tone).
2. Reference retrieval — the pipeline pulls a set of proven high-performing thumbnails matched to your platform and category, or clones your uploaded Custom Dataset if you passed a customAssetsId.
3. Image composition — a diffusion-based image model generates a fresh thumbnail conditioned on the references and the title signals. If useLogo: true, the logo is composited on top of the final image as a sharp overlay (never sent to the model, so it stays crisp).

End-to-end latency is under 25 seconds. You get back the image in the same HTTP response — no polling required.

Controls You Have

The AI does not run wild. You control the output at three levels:

• Smart References — pick the platform (format) and niche (category). The AI generates in the style of that platform/niche's proven patterns.
• Custom Datasets — upload 1–6 of your own reference thumbnails and pass customAssetsId. The AI clones your style at 80–100% fidelity.
• Compositing flags — usePhoto adds your headshot as the main subject; useLogo overlays your brand logo on the final image.

Output Quality

Two quality tiers:

• 1K (default) — model: "sd", 10 credits per generation. Available on all plans including the free tier.
• 2K (Pro+ only) — model: "hd", 20 credits per generation. Higher resolution and detail, gated to Pro and Business plans.

Output format is webp by default (~30% smaller) or png on request. Dimensions are determined by the format parameter — YouTube 1280×720, blog post 1200×630, LinkedIn 1200×627, X 1200×675, Instagram 1024×1024.

Response Format

Every generation returns both a base64 inline image and a public Cloudflare R2 CDN URL in the same JSON response. Use whichever fits your workflow — decode base64 for immediate upload to YouTube, or hand off the URL for a CDN-first pipeline.