Smart References — Auto-Discover Proven Thumbnail Patterns

Every ThumbAPI generation is guided by a live-updated reference library of high-performing thumbnails. Pick your target platform, pick a category (or let the LLM decide), and the AI generates in the style of thumbnails that already convert on that platform in that niche. Smart References is on by default. Zero config, no flag to enable.

Why "Smart" References

Generic AI image models do not know the difference between a thumbnail that gets 12% CTR on YouTube and one that fails on LinkedIn. Different platforms reward different visual patterns — bold text and shocked expressions on YouTube, minimalist typography on LinkedIn, high-contrast product shots on Instagram, punchy single-idea framing on X. A general-purpose image model gives you an average of all of them, which performs like none of them.

Smart References solves this by matching every generation to references from your platform, in your category. The AI does not guess what a "YouTube crypto thumbnail" should look like. It is shown actual high-performing YouTube crypto thumbnails and generates in that pattern.

How It Works

Send a title and a format (the target platform). Optionally pin the category — or leave it on auto and let the LLM classify your title into the closest niche. That is the whole surface.

curl -X POST https://api.thumbapi.dev/v1/generate \
  -H "x-api-key: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "title": "10 crypto scams to avoid in 2026",
    "format": "youtube",
    "category": "auto"
  }'

What each parameter does

• format (required) — which platform the thumbnail is for. Determines which reference pool the AI pulls from. Accepts youtube, instagram, linkedin, x, or blogpost. Different platforms mean different reference sets, dimensions, and style rules.
• category (optional) — which niche inside that platform. Defaults to auto, in which case the LLM reads your title and picks the closest matching category on its own.

Category Values

Set category: "auto" to let the LLM classify — or pin one of these 13 values explicitly when you want direct control:

• auto — LLM picks based on the title (default)
• tech-saas
• business-finance
• education-tutorial
• fitness-wellness
• medical-healthcare
• lifestyle-vlog
• food-cooking
• travel
• gaming
• entertainment-comedy
• news-commentary
• creative-design

The category biases both the reference search and a soft final-tone hint appended to the AI prompt. It never overrides the references — it narrows them.

What "Live-Updated" Means

ThumbAPI's reference library is not a static template collection. It is continuously refreshed with new high-performing thumbnails, so what the AI learns from stays current with what is actually converting this month, not what worked in 2023.

You do not manage the library. You do not curate it. You do not update it. That is the point — the freshness is the product. When YouTube thumbnail trends shift (bigger text, more contrast, new emotional cues), Smart References tracks that shift and future generations reflect it automatically.

Smart References vs. Custom Datasets

Smart References is the default and answers the question: what is working in my niche right now?

Custom Datasets answers a different question: what does my own brand look like? When you pass a customAssetsId, the AI clones the style from your uploaded 1–6 image set at 80–100% fidelity instead of auto-discovering from the library. Custom Datasets replace Smart References for that request — they do not layer on top.

Use Smart References when:

• You do not have a fixed brand style yet
• You want thumbnails that follow current niche patterns
• You publish across niches and want each thumbnail matched to its topic
• You are just getting started and want good-looking output on your first call

Use Custom Datasets when:

• You already have a distinctive thumbnail style you want to keep
• You run a channel with strict brand guidelines
• You want every thumbnail to look like it came from the same designer

Who This Is For

YouTube Automation Channels

Faceless channels operating across niches — crypto, finance, health, gaming — need thumbnails that look native to each niche. Smart References gives you that with a per-request category switch. One API, many channels, no per-channel style tuning.

Multi-Platform Creators

The same title generates different-looking thumbnails on YouTube, Instagram, and LinkedIn — because each platform has its own reference pool. You do not have to design three variants; you send three requests with different format values.

Agencies

Client videos span niches. Smart References handles the niche shift automatically — you do not have to remember which client is in which vertical or which template applies. Send the title, set the category, get output that fits.

Developers Building Automation Pipelines

When ThumbAPI sits inside an n8n, Make, or Zapier workflow, the person triggering the workflow is not a designer. Smart References means the pipeline produces platform-appropriate output without any human in the loop making style decisions.

Under the Hood (Short Version)

Two signals drive every generation: the reference set retrieved for your platform and category, and the tone hint derived from your title and category. The reference set constrains the visual style; the tone hint biases text, color, and mood. The AI composes a new image conditioned on both.

The reference set is refreshed continuously from a proprietary source of high-performing thumbnails — a moving target, not a fixed template library. That is why generations stay current without you doing anything.

Pricing

Smart References is included on every plan, including the free tier. There is no add-on charge and no separate credit cost for using it — it is the default behavior of every POST /generate call.

The base cost for a 1K (sd) generation is 10 credits. The free tier includes 50 credits per month, no credit card required. See full pricing for higher tiers.