sample-assets

sample-assets is a collection of sample material — data, code, images, video, audio, documents, and prose — kept in one place so that a demo, a test, a tutorial, or a prototype never stalls on the search for a file to feed it.

Every developer ends up with the same scattered pile: a CSV somewhere in Downloads, a placeholder photo borrowed from a blog post, a PDF that happened to be lying around. This repository is that pile, organized, named consistently, and free to redistribute.

The structure is in place, and the assets are being added over time. Folders you find empty today are waiting for their content, not missing it.

What's inside

Folder	Holds	Organized by
datasets/	Synthetic data for imports, fixtures, and parser tests	File format, then topic
code/	Source snippets and small self-contained programs	Programming language
images/	Photos, illustrations, and interface icons	Kind, then photo subject
videos/	Short clips for players, thumbnails, and transcoding	Flat
audio/	Sounds, tones, and loops for players and waveform tests	Flat
documents/	PDFs, office files, and other document formats	File format
text/	Long-form prose and verse as plain text, no markup	Flat, genre in the file name

Each folder carries its own README.md describing what belongs in it, which formats it accepts, and how its files are named. Read that one before adding anything.

Layout

datasets/        Sample data. One subfolder per format:
  csv/             comma-separated values
  json/            JSON and JSON Lines
  sql/             dumps and seed scripts
  tsv/             tab-separated values
  txt/             plain text, log lines, word lists
  xml/             XML documents and feeds
  yaml/            YAML configuration and data
code/            Source code samples. One subfolder per language:
  c/  cpp/  csharp/  dart/  go/  java/  javascript/  kotlin/
  php/  python/  ruby/  rust/  shell/  swift/  typescript/
images/
  photos/          photographic images, grouped by primary subject
    animals/       wildlife and other animals
    backgrounds/   scenic and atmospheric backgrounds
    food-and-drink/ food and serving scenes
    nature/        plants and natural details
    objects/       products, tools, and still lifes
    people/        fictional people on transparent backgrounds
    places/        architecture, interiors, and places
    transportation/ vehicles and other transport
  illustrations/   drawn artwork
  icons/           interface marks, preferably SVG
  objects/         single objects on transparent 250x250 canvases
videos/          Short video clips
audio/           Sound files
documents/       Documents. One subfolder per format:
  adoc/  css/  docx/  epub/  html/  md/  odp/  ods/
  odt/  org/  pdf/  pptx/  rst/  rtf/  tex/  xlsx/
text/            Long-form plain text: stories, essays, reports, poems

Datasets are indexed twice

Sample data is wanted in two different ways — "I need a CSV" and "I need a list of users" — so datasets/ answers both. Folders are formats. File names are topics. A topic keeps the same name in every format it exists in:

datasets/csv/users.csv     ┐
datasets/json/users.json   ├─ the same records, three ways
datasets/sql/users.sql     ┘

Pick the topic and you know the file name; pick the format and you know the folder. When one topic comes in several sizes, the row count goes in the name — users-100.csv, users-10000.csv — and a plain users.csv is always the small, quick-to-read one.

Three folders hold text, for three different reasons

Plain text turns up in three places, and the difference is what the file is:

datasets/txt/server-access-log.txt   data to parse
text/story-the-last-round.txt        prose to read
documents/md/readme-sample.md        a document, with markup

A log, a word list, or a fixture is data, whatever its extension. A story, a report, or an essay is prose, and lives in text/ with no markup at all. Writing that needs headings, italics, or a page layout is a document.

Using the assets

Clone the whole repository:

git clone https://github.com/jooy2/sample-assets.git

Or take only the folders you need, without downloading the rest:

git clone --filter=blob:none --sparse https://github.com/jooy2/sample-assets.git

git sparse-checkout set datasets/json images/icons

Grab a single file from the command line:

curl -O https://raw.githubusercontent.com/jooy2/sample-assets/main/datasets/json/users.json

Or link one straight into a page or a fixture, served by jsDelivr's CDN:

https://cdn.jsdelivr.net/gh/jooy2/sample-assets@main/images/photos/backgrounds/alpine-lake-dawn-1536x1024.jpg

Pin @main to a tag or a commit hash when a sample must not change under you.

Conventions

The rules are short, and each folder's README.md states its own specifics:

• Lowercase kebab-case names, describing the subject: world-cities.csv, notification-ping.mp3, icon-download.svg. Code is the exception — a sample there is named the way its own language names files, so HelloWorld.java stays HelloWorld.java.
• Qualifiers go in the name, not in a nested folder — a row count, a resolution, a duration, or the shape a sample exists to exercise: users-nested.json, mountain-lake-1920x1080.jpg, silence-5s.wav. The subject-category folders under images/photos/ are the deliberate exception.
• Deliberately broken samples say so, with -malformed, -invalid, or -broken in the name. Those are the only files exempt from the format check in CI.
• UTF-8 and LF, unless a sample exists to demonstrate the opposite.
• Files stay small. A sample is meant to be cloned, not archived.

Everything under datasets/ is checked on every push: it must parse as the format its extension claims, sit in the matching folder, and follow the naming rule. Run the same check locally before opening a pull request:

python3 .github/scripts/validate-datasets.py

Generated content

Some of the assets in this collection are produced by generative AI agents — data rows, illustrations, code samples, and document text among them. That is deliberate: a sample invented for this repository can be handed to anyone, which is not true of material borrowed from somewhere else.

None of it depicts or describes anything real. The people, faces, companies, products, places, and events in these assets are inventions rather than records. They are not based on any actual person, organization, or location, and any resemblance to a real one is coincidental.

So treat a generated asset as a stand-in, not as a source. The figures in it are not accurate, the text in it is not authoritative, and the code in it has not been reviewed for production use.

What does not belong here

• Real personal data. Sample records are invented. Names, addresses, and card numbers in this repository describe nobody.
• Credentials, live API keys, or anything scraped from a service that did not offer it.
• Assets that are not free to redistribute. If it cannot be handed to a stranger with its license attached, it cannot be here.

Licensing and attribution

The repository and the assets created for it are released under the MIT License.

An asset that comes from somewhere else keeps its own license, and is listed in its folder's README.md with the source, the author, and the terms. Check that list before shipping an asset in a product — MIT covers this collection, not necessarily every file in it.

Contributing

Anyone can contribute to the project by reporting new issues or submitting a pull request. For more information, please see CONTRIBUTING.md. Participation is subject to the Code of Conduct.

To report a security issue, or an asset that should not have been published, please follow the process described in SECURITY.md.

Author

CDGet · cdget.com · Contact

License

Please see the LICENSE file for more information about project owners, usage rights, and more.