free-ppt-template: a catalog of 595 generated PowerPoint decks

A catalog, not a repository of decks

The most important line in this README is a note, and it explains the design of the whole project.

This repository stores no binaries. It is a catalog: the decks and their preview images are hosted on the project's own CDN, at cdn.ppttemplate.ai, so cloning costs a few megabytes instead of gigabytes.

That is a design decision rather than a limitation, and it has a clear reason. A collection of 595 decks with 6,603 slides in them is measured in gigabytes if you keep the .pptx files in git, and every clone, every fork, and every CI job pays for that. Splitting the catalogue from the artefacts means the repository stays small enough to read and the files stay on a CDN that can serve them directly.

So the repository contains four things. The README, the LICENSE, a categories directory with one Markdown file per category, and index.json, the machine readable index of the whole catalogue. The live catalogue itself is at ppttemplate.ai, and every download link in the README is a direct link to a .pptx.

The scale, as the README states it, is 595 free AI-generated presentation templates, 6,603 slides at roughly 11 per deck, 19 categories, and the format is .pptx in 16:9 with a preview image of every slide. The project also states that there are no stock photos, no watermarks, no signup, and no account.

One inconsistency is worth flagging for anyone reading the repository description rather than the README. The description advertises more than 4,000 templates, while the README and the category table both say 595, and the category counts add up to exactly 595. Trust the README and the table, and expect the description to lag.

The category map, and where the balance sits

The category table is the most useful navigation in the project, and it is also a map of what the generator was actually asked to produce most often.

Education and Classroom is the largest category by a wide margin, with 149 templates, covering lesson plans, classroom activities, and teaching resources. General Purpose has 41, described as flexible decks that fit almost any topic, while Business and Corporate has 55. Data, Charts and Infographics has 43, Thesis and Academic has 40, Pitch Deck has 38, Medical and Health has 36, and Aesthetic and Minimal has 34.

The long tail is instructive. Marketing and Social has 30, Project and Planning 29, Sales and Revenue 25, Holidays and Events 18, Creative and Portfolio 18, Finance and Accounting 16, Kids and Playful 11, Nature and Environment 5, Documents and Spreadsheets 3, and then two categories with two templates each, Technology and Product and Food and Travel.

That distribution tells you something about the generation pipeline. A category like Education and Classroom is full of layouts that share a structure, a classroom board, a lesson plan grid, a worksheet, so the same visual system can be reused with different copy. A category like Technology and Product, at two templates, is the opposite case, where a software product deck has no obvious repeatable layout and the generator had to invent one per request.

Each category has its own Markdown file under categories, which is where the description of the category and the list of its templates live. That is the file to read if you want to know what is in a category before you download anything, and it is the file a pull request would touch when a template is missing from a category.

Browsing by colour, and the numbers behind the index

There is a second navigation axis, colour, and it is listed with counts. Blue leads with 174 templates, followed by Colorful with 136, Green with 116, Black with 115, Orange with 89, White with 89, Pink with 86, Yellow with 78, Red with 66, Beige with 65, Pastel with 64, Purple with 61, Grey with 54, Brown with 29, Violet with 17, Gold with 16, Gray with 9, and then the tail: Monochrome with 5, Navy with 4, Cream with 3, Multicolor with 3, Teal with 3, Rainbow with 2, Coral with 1, and Silver with 1.

A few things fall out of that list. Grey and Gray are separate entries, which is a data hygiene problem rather than a design distinction, since a template that is grey gets counted in one bucket and a template that is gray in the other. The same kind of duplication probably exists elsewhere in the tags.

The distribution also shows that the palette is not random. The top of the list is dominated by colours that are legible as a background for text, which is what a slide generator would reach for, and the two colours that appear once, Coral and Silver, are the outliers.

For a designer, the useful reading is that the catalogue is heavy on blue, green, and black decks, and light on anything unusual. If you want a deck in a specific colour, the practical approach is to search the index for the tag rather than to browse the colour page, because the tag list is what the scriptable interface uses.

The colour counts are also a useful sanity check on the index itself, since they are presumably derived from the same tags that the bulk download examples filter on.

How a deck is made, and what that means for the file you get

The generation pipeline is described in a short paragraph, and it explains the character of the output.

Each deck is generated end to end: a language model writes the title, structure, and copy; an image model renders every slide as a full-bleed 16:9 design; the slides are packaged into an editable .pptx. Nothing is traced or re-hosted from a stock library, and no deck carries a watermark, logo, or attribution mark.

The order matters. The copy is written first, by a language model, which means the text on the slides is generated text with the usual properties of generated text, and the slide is then rendered as a picture of that text. The full-bleed description is the important detail: each slide is a rendered image, which is why the decks can look consistent in a way that a set of layout boxes rarely does, and also why the text in them may occasionally be garbled.

The .pptx packaging then puts those images into a file, and the file is editable in the sense that you can open it, reorder slides, and add your own, but the individual slide content is an image rather than a set of native text boxes and shapes.

That has consequences in both directions. You get a deck that opens in any tool that handles .pptx, with a visual quality that most free template sites cannot offer, and you give up the ability to restyle a slide by editing a heading. For a talk where the design matters and the content is yours, that is a good trade. For a document that has to follow a corporate template with real text objects, it is not.

The formats it opens in are listed explicitly: PowerPoint on Windows, Mac, and Web, Google Slides, Keynote, LibreOffice Impress, WPS Office, OnlyOffice, and Zoho Show. For the first three, the instruction is to download the .pptx and open it, and for Google Slides it is File, then Import slides, then Upload, and then pick the .pptx.

Scripting the catalogue with index.json

index.json is the part of the repository that makes it more than a list of links. It carries every template with its title, description, tags, colors, the detail page, and a direct .pptx URL, and the README says that means the whole catalogue is scriptable.

The three examples given in the README show what that means in practice, and they are worth copying rather than rewriting, because they show the shape of the data.

# grab every minimalist business deck
jq -r '.[] | select(.tags | index("Business")) | select(.tags | index("Minimalist")) | .pptx_url' index.json | xargs -n1 -P4 curl -O

# list decks with at least 12 slides
jq -r '.[] | select(.slide_count >= 12) | .title' index.json

# every blue template's detail page
jq -r '.[] | select(.colors | index("blue")) | .url' index.json

The first one shows the two-axis selection that the colour list and the tags imply, and the download itself is parallelised with xargs, one file at a time and four at once, which is the right shape for a CDN-hosted artefact set.

The second one shows that there is a slide_count field, which is a useful filter, because a deck with eleven slides and a deck with twenty have very different uses, and the second example prints titles rather than downloading, which is the fast way to find out what exists before spending bandwidth on it.

The third shows that the colour dimension is a separate field from the tags, and that the values are lower case, which is a small trap for anyone writing a filter of their own.

Together they show the whole workflow: query the index with jq, decide what you want, and fetch the .pptx URLs directly. Nothing in that path needs the website.

Contributing, and the regeneration loop

The contribution guide is short and points at three kinds of help, and the first one is the interesting one.

Bad renders: open an issue with the template slug, and garbled text and broken layouts get regenerated. That is a real feedback loop rather than a complaint queue, and it is possible because the decks are generated artifacts with a slug that identifies them, so a report can be turned into a regeneration request for exactly that template.

Missing topics: tell them what you searched for and could not find, which is the other way the catalogue grows.

Better metadata: tag and description fixes are easy pull requests against index.json, which is the lowest friction contribution available and, given the Grey and Gray duplication in the colour list, the most immediately useful one.

The licence is CC BY 4.0, stated in the README table as free for personal and commercial use, with a LICENSE file at the root. That is worth understanding properly before you deploy a deck in a commercial context, because a Creative Commons attribution licence carries obligations that a template you found on a slide sharing site usually does not, and the classification GitHub shows for this repository is a custom licence that it cannot be classified automatically, so the file is the authority.

On the state of the project, it is not archived and the last push was on 2026-08-31, with no GitHub releases, which is consistent with a catalogue that is updated by pushing new entries and regenerating broken ones. The content is the asset here, and it is served from the catalogue site rather than from the repository, so the repository is a thin, scriptable index over a larger hosted collection.

Editorial conclusion

free-ppt-template fits someone who needs a deck quickly, in any of the listed formats, and who is willing to treat the generated design as a starting point rather than a finished artefact, since the copy and the layout both come out of a model. It does not fit a user who needs editable, native slide objects in a specific corporate template, because the slides are full-bleed rendered images packaged into a .pptx. Verify first that the licence, CC BY 4.0, matches how you intend to use the decks, that the file you download opens in the tool you use, and that the text is legible and correctly spelled, because the project asks you to report garbled renders and regenerates them.

Frequently asked questions

Official sources

1. ai-ppt-template/free-ppt-template on GitHub
2. Issues
3. Project website
4. README