Features

Fixtures you can reason about, because a script made every one of them.

• Generated, never collected

  All 5,075 files come out of a deterministic pipeline with fixed seeds. That is why every file has an exact specification and why the whole library is copyright-clean.

• 27 categories, 288 extensions

  Images, audio, documents, data, video, archives, e-books, 3D models, fonts, email, AI/ML, source code, web assets, localization, security, testing and QA reports, version control, observability, supply chain, geospatial, scientific, pipelines, templates, and forms.

• 202 documented pairs

  Clean and noisy, colour and greyscale, source and scanned, valid and truncated. Pairs link to each other, so before-and-after testing needs no setup.

• Deliberately broken files, clearly labelled

  Truncated PDFs, invalid JSON, ragged CSVs, malformed markup. Every one says “intentionally corrupt” in its title and description so it can never be mistaken for a healthy sample.

• Permanent URLs

  Slugs, file ids, and download paths never change once published. A fixture referenced from your CI config keeps working.

The page is the documentation. Nothing important is hidden behind a download.

• An exact spec sheet

  Dimensions, encoding, sample rate, bit depth, line endings, form field names, noise sigma, silence timestamps, seed — whatever actually defines that file.

• An inline preview

  Images, audio with a waveform, video, 3D models, text and data rendered in place. Files that cannot be previewed honestly — encrypted, empty, or intentionally corrupt — get an information card instead of a player that can never load.

• One download button

  Exactly one per page, never adjacent to an advert, and served with a forced-download header so nothing renders in place unexpectedly.

• Pair, group, and related links

  The counterpart file, the conversion group it belongs to, and files that exercise the same tooling.

• Purpose tags and a convert hand-off

  Each file is tagged with the testing purposes it serves — 131 of them across the library — and links straight into the sister converter pre-set to its format.

Five different routes in, because people arrive knowing different things.

• Browse everything

  The full directory, filterable by category, subcategory, format, and purpose.

• Search

  Fuzzy search over files, templates, forms, guides, and pages, running entirely in your browser against a prebuilt index. No search server, no query logging.

• Browse by purpose (131)

  Start from what you are testing — OCR, denoise, auto-trim, CSV parsing, model inference — rather than from a file type.

• Format coverage matrix

  331 formats available today, 1 coming, and 11 we deliberately do not support — each with the reason why.

• Format glossary (289 formats)

  Plain-language explanations of what each format is and how to test with it, each linking to real downloadable examples.

Six workspaces. All of them run on your device — nothing is uploaded, ever.

• Choose a workspace

  A single entry point that routes you to the right editor or synthetic-media lab for the file you need.

• Text and data editor

  Open any text-based example or template with syntax highlighting for JSON, YAML, XML, CSS, HTML, SQL, Markdown, JavaScript and more, edit, and download the result.

• Form Studio

  Build a form from scratch or start from one of the 243 catalog form definitions, preview it live, and export standalone HTML, JSON, or a fillable PDF.

• Document Studio

  Start blank or from one of 36 editable document families and export DOCX, XLSX, HTML, Markdown, or print-ready PDF.

• Visual Template Studio

  Replace product images, edit text and colours on a real scene graph, and export PNG, JPEG, or WebP locally.

• Audio creation lab

  Synthesize tones, sweeps, seeded noise, or silence and export bounded PCM or encoded fixtures with versioned recipes and SHA-256 evidence.

• Video creation lab

  Render geometric test patterns and encode only the containers and codecs this browser session reports, without camera or microphone access.

Usable starting points rather than test fixtures — the one part of the site that is not about breaking things.

• 840 visual templates across 42 categories

  Product photography, marketplace listings, social posts, ads, and marketing scenes. Every template is composed from a scene graph, so the committed preview and the Studio canvas cannot drift apart.

• 43 document families

  Invoices, statements of work, press releases, pitch decks, meeting notes, OKRs, contracts, CVs, and HTML email, generated with documented fields and formulas.

• Legal markings that cannot be deleted

  SAMPLE markers, fictitious issuers, and category disclaimers are locked nodes in the scene, not prose — so an edit cannot quietly strip them.

• OBS-compatible scene collections

  Selected visual templates ship a scene collection alongside their image exports, for OBS-compatible streaming software. Asset filenames are matched to the download names on purpose, so a missing-file re-link is one click rather than source by source. Not affiliated with the OBS Project.

Pages that exist to be automated against, not downloaded.

• Live form demos

  Contact, full-contact, signup, survey, job, checkout, and upload flows with real client-side validation and copyable source — targets for scrapers, autofill tools, and form parsers.

• A demo endpoint that stores nothing

  Submissions hit one POST route that validates and discards. It is same-origin-only, rate-limited, honeypot-guarded, and byte-capped.

• 243 downloadable form fixtures

  HTML and fillable PDF AcroForms covering employment, housing, donation, enrollment, RSVP, tax, KYC, shipping, insurance, NDA, consent, loan, and payment cases — all with fictional data, and many paired across formats.

• Live accessibility and scraper targets

  ARIA landmarks, nested modal focus stacking, infinite scroll, sortable and filterable tables, file-drop zones, nested iframes, scroll-spy, forced light and dark themes, print stylesheets, slow loads, and a cookie banner.

Written to be useful on its own, not as SEO filler around a download button.

• 39 test-recipe guides

  Practical walkthroughs — testing a denoise filter, CSV edge cases that break parsers, loudness and peak-clip testing, ONNX inference contracts — each with a bespoke hero image.

• Documentation

  How spec sheets, pairs, conversion groups, purpose tags, licensing, and CI usage actually work.

• Tutorial and FAQ

  A guided tour of browsing, searching, downloading, and creating, plus direct answers on licensing, safety, and requests.

The library is meant to be consumed by tooling and answer engines, not only by people.

• Full sitemap

  Every indexable URL, dated from the content that backs it rather than the build clock, so a code-only deploy does not tell crawlers that everything changed.

• llms.txt

  A concise, link-rich Markdown summary generated from the live catalog, so assistants can read the library instead of scraping it.

• Structured data

  Organization, WebSite, BreadcrumbList, and per-file JSON-LD emitted server-side on every relevant page.

• Bespoke social cards

  Almost every route generates its own 1200 x 630 image at build time, including every file page, template, and guide, rather than reusing one shared card.

• Standards files

  robots.txt that explicitly welcomes AI crawlers, a web app manifest, ads.txt, and an RFC 9116 security.txt.

The platform decisions that make the rest of it trustworthy.

• Free, with no account

  No sign-up, no login, no paywall, no rate-limited free tier. The site is funded by advertising and the files are free for any use with no attribution required.

• Nothing you make leaves your device

  Every editor and Studio works locally. There is no upload endpoint, no database, and no server-side secret anywhere in the deployment.

• Region-aware consent

  Opt-in where prior consent is required, opt-out elsewhere, and Global Privacy Control forces the strict default everywhere. Every cookie is listed by name.

• Installable and offline-capable

  A progressive web app with a cached shell, so previously visited pages still render with no network.

• Light and dark, keyboard-first

  A skip link, visible focus rings, labelled landmarks, and full reduced-motion support — with the measured contrast ratios and the open gaps published rather than glossed over.