Free Stock Photo APIs Compared: Unsplash, Pexels, Pixabay, Openverse & Pexafy

You need free-to-use photos inside a product. The obvious move is to grab an API key from Unsplash, Pexels or Pixabay and ship. Two weeks later you discover the part nobody compares: your app is stuck at 50 requests an hour until a human approves it, one provider forbids hotlinking, another wants a tracking call on every download — and your search results are only as good as the tags a stranger typed years ago.

This page is the comparison we wanted when we started building Pexafy: quotas in the same unit, the terms that actually bite in production, and what each API can and cannot do. Every competitor number comes from the provider's own docs, linked at the bottom. We also say plainly where Pexafy is not the right answer.

The comparison table

Five options, fourteen criteria. Unsplash, Pexels and Pixabay are single libraries with their own APIs. Openverse and Pexafy are aggregators — they search many libraries at once — which is the distinction most comparison articles quietly skip.

Swipe the table sideways to see every provider →

Sources: each provider's public documentation, checked 14 August 2026 — links in the references at the end. Found something out of date? Tell us and we'll correct it.

Quotas: read the unit, not the number

Quota comparisons are where these articles usually go wrong, because the four APIs count four different things over four different windows. Converted to a common unit — a search a user actually sees — the ranking changes:

Be honest about what this says: if you need raw volume against a single library, Pixabay's 100 requests per minute is the most generous free tier on this page, and we won't pretend otherwise. What it doesn't give you is coverage. To put the same result set together yourself from three libraries, the cost isn't one request — it's three, plus everything around them:

The unit that matters. One Pexafy request = one semantic search across 9 libraries, deduplicated and ranked in a single list. With the providers' own APIs, the same user-facing search costs 3 to 5 calls plus the merge layer — code you write, test, and pay for in maintenance forever.

The rules you discover after you ship

Quotas are advertised. Terms are not. These are the four clauses that most often force a rewrite — all of them public, all of them easy to miss:

1. Unsplash: production access is a human decision. Apps start in demo mode at 50 requests/hour. To be raised, you submit screenshots, a title and a description, and your app must demonstrably follow the guidelines — including view and download reporting. Until someone approves it, 50/hour is your ceiling, in production, with real users.
2. Unsplash: the download-tracking call. Every time your app downloads a photo, you must also call the /photos/:id/download endpoint. It's not optional, it's part of the guidelines, and it's a second round-trip you have to remember in every code path that saves an image.
3. Pixabay: no permanent hotlinking, and cache for 24 hours. You must download images to your own server for permanent use, and API responses must be cached for 24 hours. That means storage, a cache layer and an invalidation policy — infrastructure, before you've shown a single photo.
4. Everyone: attribution rules differ from licence rules. The photo licences are famously permissive, but the API terms add their own obligations — a prominent link back to Pexels, credit to the photographer and Unsplash, showing users where Pixabay images come from. Read the API terms, not just the licence page.

Pexafy's position, stated plainly: your key works on the first call, there is no application review, no mandatory tracking endpoint, and no caching obligation. Attribution isn't required by our API terms — and every result still ships with a ready-made attribution string (HTML and plain text), because crediting photographers is good practice and we'd rather make it one line of code than a chore. The source library's own licence, of course, always applies.

Keyword vs semantic: what changes in the results

Every API on this page except Pexafy matches your words against words attached to a photo — tags, titles, descriptions. Pexafy compares meaning: your sentence and every image are turned into numerical fingerprints, and the closest ones come back. Two consequences show up immediately in production.

First: coverage you didn't ask for. Here is one real Pexafy request — one query, one ranked list. Look at the source badges:

a lone tree standing in an empty green field under a wide blue sky

Top resultPicjumbo Pexels Pixabay Unsplash Pixabay Unsplash

Second: the query stops being a keyword puzzle. With tag matching, every extra word is one more thing the tags probably don't contain, so you learn to type tree field and scroll. With semantic search the opposite holds — “a lone tree standing in an empty green field under a wide blue sky” is a better query than “tree”, because every detail sharpens the match. Same effect for moods (calm, nostalgic, minimal) and for languages: the sentence lands in the same place whether you write it in English, French or Japanese, against the same single index. We took that apart with real searches in semantic image search vs keywords.

One schema instead of three

This is the part that costs the most engineering time and never appears in a comparison table. The same concept — “the medium-size image URL” and “who took it” — is named differently in every API:

// Unsplash
results[].urls.regular        results[].user.name
// Pexels
photos[].src.large            photos[].photographer
// Pixabay
hits[].webformatURL           hits[].user

Multiply that by dimensions, orientation, dominant colour, source page URL, licence and pagination, and you get an adapter per provider, a normalisation layer, a dedupe pass (the same photo is often on several libraries) and a re-ranking heuristic — typically a few hundred lines that must be kept in step with three changelogs. With Pexafy that layer is the product:

curl -H "X-Api-Key: $PEXAFY_KEY" \
  "https://api.pexafy.com/api/v1/search/photos?q=a+lone+tree+in+an+empty+green+field"

{
  "success": true,
  "data": [
    {
      "photo_id": "019e4cd9-6369-70c3-ab2b-0ffd7e7965ad",
      "urls": { "thumb": "…", "small": "…", "regular": "…", "large": "…", "full": "…" },
      "width": 5376, "height": 3584, "orientation": "landscape",
      "color_name": "xanadu", "color_hex": "#697E77",
      "photographer_full_name": "Viktor Hanacek",
      "photographer_url": "https://picjumbo.com/about-viktor-hanacek/",
      "source": "Picjumbo", "license_type": "free",
      "relevance_score": 0.80,
      "alt_description": "A single tree standing in a wide green field",
      "attribution": { "html": "Photo by …", "plain": "Photo by …" }
    }
  ],
  "meta": { "took_ms": 147 }
}

Identical fields for a Picjumbo photo, an Unsplash photo and a Pixabay photo. One relevance_score comparable across every source, because they were ranked together rather than stitched. And an attribution block you can render as-is.

Two things none of the others do

1 · Search by image, not just by words

Give Pexafy an image — a URL, or one you upload — and it returns visually and semantically similar photos across every indexed library. You can steer it with text at the same time (“like this photo, but at sunset”). It's how you build “more like this”, or how you find a free alternative to an image you can't license. None of Unsplash, Pexels, Pixabay or Openverse exposes an image-to-image endpoint.

2 · An MCP server, so AI agents can search images natively

Pexafy runs a hosted Model Context Protocol server at mcp.pexafy.com/mcp with OAuth 2.1 — connect it to Claude, or to your own agent, and the model searches 9 photo libraries by itself, sees the thumbnails inline, and hands back credited images. Three tools: text search, search-by-image, and similar-photos. The other providers have no official MCP server; what exists are community wrappers you host and key yourself. If you are building an agent that needs pictures, that difference is the whole integration.

Which one should you actually use?

A comparison where one product wins every row is marketing, not information. Here is the honest split:

• Pixabay — you need high-volume, single-library access on a free tier and you can host the images yourself. 100 requests/minute is unmatched here.
• Unsplash — the visual identity of your product depends on Unsplash's editorial style specifically, and you're willing to go through app review and implement the tracking call.
• Pexels — you want one clean library with an instant key, a generous hourly rate, and no infrastructure obligations.
• Openverse — you need Creative Commons and public-domain material beyond the commercial stock libraries, including museum and institutional collections.
• Pexafy — you want breadth without integration work: one key, 9 libraries, one schema, meaning-based ranking, image-to-image search, and an MCP server for AI agents. Or you're an agency, a CMS, or an AI product where “find the right picture” is a feature, not a page.

You can also use both: a lot of teams keep a direct Unsplash integration for brand photography and use Pexafy for everything else, precisely because it costs one key and no approval to try.

References & footnotes