Make an AI-Generated PPT Editable Again
You asked an AI presentation tool for slides, downloaded the .pptx, opened it in PowerPoint - and discovered you can’t click a single word. Every slide is one flattened picture with the text baked into it. Upload that file here and get it back with the design intact and real, editable text boxes rebuilt on top.
What this can and can’t do
Your original slide picture stays as the slide background, so the design - colours, icons, charts, layout - survives exactly. The text is then read off that picture and rebuilt as real, clickable PowerPoint text boxes on top of it. That means it’s a best-effort reconstruction, not a pixel-perfect vector rebuild: expect the odd misread character, a line that’s split or merged oddly, or a font size a point or two off, and the original font name can’t be recovered from an image at all. Read through the result before you present it.
Why your AI deck isn’t editable
A .pptx is a zip of XML. A real PowerPoint slide stores each text box as a shape with its own <a:t> text runs, so PowerPoint knows where every word is and lets you retype it. Several AI presentation generators skip that entirely: they render the finished slide to a PNG or JPEG in the browser and drop that one picture into the slide, full-bleed. The file opens perfectly, prints perfectly - and contains no text at all. Nothing is corrupted and there is no repair option to click, because as far as the format is concerned the file is exactly what it claims to be: a picture.
This tool goes the other way. It looks at each slide’s real shape tree, finds the ones that are a single picture covering the whole slide with no text of their own, reads the words and their positions off that picture, and writes them back as genuine text shapes - keeping the picture underneath so the design is untouched. Slides that already contain real text are copied straight across and never touched.
What you get back
| Part of the slide | What happens to it |
|---|---|
| Headings and body text | Read back with OCR and rebuilt as real, selectable text boxes you can retype |
| Text position and size | Placed where the words sit in the image, at a point size estimated from their height |
| Text colour | Sampled from the picture itself, so light text on a dark slide stays light |
| Background, colours, icons, charts | Kept exactly - the original image stays as the slide background |
| Font family | Not recoverable from an image; the deck's default font is used |
| Slides that already had real text | Copied through unchanged - no OCR is run on them at all |
It is a best-effort de-flattening rather than a perfect reconstruction: the recovered text sits on top of the original design image, so OCR mistakes and slightly imprecise line positions are both possible. Read through the result before you present it.
Questions fréquentes
Why can't I edit the text in my Gamma / Copilot / Canva PowerPoint export?
Because there is no text in it. Those exports commonly render each finished slide to a single image and put that image on the slide full-bleed, so the headline, the bullets and the numbers are all pixels rather than PowerPoint text objects. PowerPoint will happily open and print the file, but there is nothing to click into. That is also why 'Repair', 'Reset slide' and re-saving the file never help - nothing is broken, the text simply isn't there.
How does this tool get the text back?
It unzips the .pptx and reads each slide's real shape tree. A slide counts as flattened when one picture covers about 85% or more of the slide and there is essentially no real text on it already. For those slides only, the background picture is run through OCR to find every word and its position in the image, those pixel positions are converted to points using the picture's own size on the slide, and each line (or paragraph, where lines are evenly spaced) is written back as a real PowerPoint text box at that position. The original picture stays as the slide background, so the design is preserved.
Will it change the slides that were already fine?
No. Slides that already contain real text or shapes are detected as such and copied across with their text, positions, font sizes, bold/italic and colours intact. OCR never runs on them, so there is no way for it to introduce a misread word into a slide that didn't need fixing. The per-slide report after conversion tells you exactly which slides were rebuilt and which were left alone.
How accurate is it?
Good enough to edit from, not good enough to publish unread. Everything comes from OCR on a rendered image, so expect the occasional misread character, a line that gets split or merged differently from the original, and a font size that is a point or two off. Text positions are placed to land on top of the original words but can drift slightly. The font family can't be recovered from a picture at all, so the deck's default font is used. Skim every rebuilt slide before you present it.
Why can I still see the old text behind the new text?
Because the original design image is still there - that's what keeps your colours, icons and layout. When 'Cover the baked-in text' is on, the tool samples the colour immediately around each line; where that area is a flat colour it paints that colour behind the new text box, which hides the old pixels completely. On a photo, a gradient or a busy area it deliberately leaves the picture alone rather than stamping a wrong-coloured rectangle over your artwork, so some ghosting can remain there. Nudging the new text box a pixel or two in PowerPoint fixes any that's left.
Is my presentation uploaded to a server?
No. The file is unzipped, OCR'd and rebuilt entirely in your browser using jszip, tesseract.js and pptxgenjs. Nothing is sent anywhere, there is no account, and nothing is stored. Large decks take a while for the same reason - the OCR is running on your own machine.
Does it work for Tome, Beautiful.ai, Decktopus or SlidesAI exports too?
It doesn't look at which tool made the file - it looks at whether a slide is one full-bleed picture with no text on it. Any .pptx with that shape is rebuilt, whichever generator produced it, including decks that were exported to images and reassembled, or slides someone pasted in as screenshots. If a slide is flattened in a different way - for example one big picture plus a couple of real text boxes - it is left alone, because it already has editable content.