I’ve done a lot of work with the Office file formats over the years, flipping to .zip and messing with the format manually. Lately, because I’ve made courses with 500+ PowerPoint slides with lots of morph transitions, I iterated to find an approach that generally worked with Claude but is never perfect (using the Claude for PowerPoint plugin).
I switched to https://bento.page/ when it showed up here on HN and that’s been much more accurate I think because HTML and CSS is even more represented in training data.
OOXML is incredibly complicated, but LLMs can write it well enough. One of my colleagues was using an LLM to generate PNG slides for a presentation. They looked pretty, but changing anything was a pain. I fed the images to free ChatGPT and asked for a PPTX with every text box, frame and image split into its own object, and it spat out a real, editable presentation.
You could probably then copy-and-paste those into a PPTX file created by PowerPoint to make sure there were no weird things going on with the file(structure) that might cause problems later.
> A PowerPoint file is a ZIP package containing XML files and binary assets.
> It stores all the properties you set in a PowerPoint presentation, and tries to organize them into separate files to avoid repetition and make things easier to locate.
This is a great read. One of the most requested AI features I’ve seen is PPTX slide generation that matches a corporate template and assets perfectly. Claude with skills comes close but not close enough. The amount of fiddling needed to adjust is nearly equal to doing it yourself from scratch. It’s a very tricky standard to work with!
I knew a powerpoint is basically a zip file when I was trying to crack a pptx file last year, the AI basically said “pptx file is basically a zip file, let’s do …”, when I unzipped it, turned out the password protection is useless, all assets and everything are in there unprotected, including the 3D models inside. And you can pack it to a file again, just like the protected one but with no password.
Why “obviously”? The file could contain two parts, one encrypted PowerPoint part with all data and original assets, and the latter just the “rendered” version with everything pruned down to what is needed to display the content. Pair this with some hashing which is signed/stuffed in the encrypted part and any PowerPoint copy would be able to detect tampering as well. Seems obviously trivial.
It's non-trivial to determine "what is needed to display the content". The closest you have is something like "print/save to pdf", which is too lossy for some (eg. you lose animations and comments), and isn't 100% foolproof either (eg. if you use rectangle tool to redact some text, you can still recover what's underneath).
i've been struggling a lot with my team to standardize our presentation systems. as soon as we got claude people just started making adhoc, non-standard, slides. This is good because they're 10x better than the bullshit one offs PMs used to make. But as a result we've got 80 different ways of conveying the same type of information to our leadership. And its pissing those leaders off.
Getting people aligned on a single template is our biggest headache for the ops side of the house. Will definitely need to research this further. Thanks for the post.
You might find my open source project SmallDocs useful. It lets Claude make presentations in Markdown which can be exported to PPT. You can define styles and make them defaults. (You can also share the native SmallDoc presentations by just sending a url.)
Oh man the masters and layout proliferation is the biggest pain.
If you can, I'd do a bare bones starting template with like Cover, section divider, and title only layouts. Lock in slide dimensions, fonts, colors. Then ask (and pray) people upload that instead of letting Claude work from scratch.
Even if everyone goes in a different direction, it's much easier to merge and conform styles if you're working on the same template with minimal placeholders.
Ideally you get a library of "best practice" slides that people use as a starting point, but realistically people love doing their thing. Key is to set yourself up for an easy cleanup.
all files are either text, binary blobs or archives (zip), no exception
I personally knew about this when I needed to extract all images from a Word document in 2015, I had the idea to rename the extension to .zip and wow a conveniently placed media folder !
since then I renamed countless file extensions to zip, just to test. it worked nearly every single time
Software like 7-Zip supports many formats, even ones that are not usually treated as archives, and recognises them by magic bytes rather than extension. You can for example add a zip extension an EXE file, and 7-Zip will open it.
A much more robust way to figure out what's in a file is using libmagic, accessible through the `file` command on *nix. I don't know if there's an alternative Windows though.
Hah! That's also how I first learned about the zip trick with pptx.
Had never considered it generalizes outside of OOXML files. I'll give it a try next time I'm poking around another file format
A lot of of the library-as-a-service startups doing excel/PowerPoint/PDF editing libs are gonna die with how ubiquitous and powerful modern LLMs are. I expect a lot of such companies to write sob blog posts similar to what the tailwind founder did about having to lay off most of their team and reinvent themselves for the agentic era. It's a sad time to be a dev right now.
I’ve done a lot of work with the Office file formats over the years, flipping to .zip and messing with the format manually. Lately, because I’ve made courses with 500+ PowerPoint slides with lots of morph transitions, I iterated to find an approach that generally worked with Claude but is never perfect (using the Claude for PowerPoint plugin).
I switched to https://bento.page/ when it showed up here on HN and that’s been much more accurate I think because HTML and CSS is even more represented in training data.
OOXML is incredibly complicated, but LLMs can write it well enough. One of my colleagues was using an LLM to generate PNG slides for a presentation. They looked pretty, but changing anything was a pain. I fed the images to free ChatGPT and asked for a PPTX with every text box, frame and image split into its own object, and it spat out a real, editable presentation.
You could probably then copy-and-paste those into a PPTX file created by PowerPoint to make sure there were no weird things going on with the file(structure) that might cause problems later.
> A PowerPoint file is a ZIP package containing XML files and binary assets.
> It stores all the properties you set in a PowerPoint presentation, and tries to organize them into separate files to avoid repetition and make things easier to locate.
Very informative read overall.
Thanks! Glad it was useful.
This is all MS Office file formats ending in “x”; xlsx, docx, pptx, etc
I had to programmatically generate PPT files a while back. Don’t recommend. Maybe better with LLMs now
This is a great read. One of the most requested AI features I’ve seen is PPTX slide generation that matches a corporate template and assets perfectly. Claude with skills comes close but not close enough. The amount of fiddling needed to adjust is nearly equal to doing it yourself from scratch. It’s a very tricky standard to work with!
I knew a powerpoint is basically a zip file when I was trying to crack a pptx file last year, the AI basically said “pptx file is basically a zip file, let’s do …”, when I unzipped it, turned out the password protection is useless, all assets and everything are in there unprotected, including the 3D models inside. And you can pack it to a file again, just like the protected one but with no password.
>turned out the password protection is useless, all assets and everything are in there unprotected, including the 3D models inside
That's only true for the edit protection password, which obviously can't be implemented in a foolproof way, and therefore it's an honor system.
https://en.wikipedia.org/wiki/Microsoft_Office_password_prot...
Why “obviously”? The file could contain two parts, one encrypted PowerPoint part with all data and original assets, and the latter just the “rendered” version with everything pruned down to what is needed to display the content. Pair this with some hashing which is signed/stuffed in the encrypted part and any PowerPoint copy would be able to detect tampering as well. Seems obviously trivial.
It's non-trivial to determine "what is needed to display the content". The closest you have is something like "print/save to pdf", which is too lossy for some (eg. you lose animations and comments), and isn't 100% foolproof either (eg. if you use rectangle tool to redact some text, you can still recover what's underneath).
is password also included along with the assets?
i've been struggling a lot with my team to standardize our presentation systems. as soon as we got claude people just started making adhoc, non-standard, slides. This is good because they're 10x better than the bullshit one offs PMs used to make. But as a result we've got 80 different ways of conveying the same type of information to our leadership. And its pissing those leaders off.
Getting people aligned on a single template is our biggest headache for the ops side of the house. Will definitely need to research this further. Thanks for the post.
You might find my open source project SmallDocs useful. It lets Claude make presentations in Markdown which can be exported to PPT. You can define styles and make them defaults. (You can also share the native SmallDoc presentations by just sending a url.)
Homepage: https://smalldocs.org/
Example presentations:
https://smalldocs.org/s/mTBw58ut_eTAIlGkYePw6-#k=4MotlKf-sLc...
https://smalldocs.org/s/cGRzzCv7LA9Kd8BxZ4bBAP#k=B7_NNXPbpLw...
Code: https://github.com/espressoplease/smalldocs
Oh man the masters and layout proliferation is the biggest pain.
If you can, I'd do a bare bones starting template with like Cover, section divider, and title only layouts. Lock in slide dimensions, fonts, colors. Then ask (and pray) people upload that instead of letting Claude work from scratch.
Even if everyone goes in a different direction, it's much easier to merge and conform styles if you're working on the same template with minimal placeholders.
Ideally you get a library of "best practice" slides that people use as a starting point, but realistically people love doing their thing. Key is to set yourself up for an easy cleanup.
Cool post. Having text animations on a blog post is REALLY distracting for me tho.
all files are either text, binary blobs or archives (zip), no exception
I personally knew about this when I needed to extract all images from a Word document in 2015, I had the idea to rename the extension to .zip and wow a conveniently placed media folder !
since then I renamed countless file extensions to zip, just to test. it worked nearly every single time
Software like 7-Zip supports many formats, even ones that are not usually treated as archives, and recognises them by magic bytes rather than extension. You can for example add a zip extension an EXE file, and 7-Zip will open it.
A much more robust way to figure out what's in a file is using libmagic, accessible through the `file` command on *nix. I don't know if there's an alternative Windows though.
TrID: https://www.mark0.net/soft-trid-e.html
If you're including binary blobs as one of your options, then both text and archives are just types of binary blobs :)
Hah! That's also how I first learned about the zip trick with pptx. Had never considered it generalizes outside of OOXML files. I'll give it a try next time I'm poking around another file format
You’d be surprised how many file formats are some flavor of “zip file containing files”.
JAR/WAR, iOS apps, and others
new PPT: Zip files old PPT: OLE pseudo FS with objects embedded.
A lot of of the library-as-a-service startups doing excel/PowerPoint/PDF editing libs are gonna die with how ubiquitous and powerful modern LLMs are. I expect a lot of such companies to write sob blog posts similar to what the tailwind founder did about having to lay off most of their team and reinvent themselves for the agentic era. It's a sad time to be a dev right now.