PowerPoint media extractor returns the embedded original

Assets come out at the resolution they were dropped in at, not the display size on the slide; PowerPoint has already renamed them image1.png / media2.mp4, so order follows add-time, not slide number.

PowerPoint Media Extractor unpacks ppt/media/ from a .pptx in your tab. Scaling a picture on a slide changes only how large it looks. The dropped-in file stays in the package at its original pixel count. This tool returns that file. It does not screenshot the slide or invent a sharper original.

Display size is a scale, not the file

A 4000-pixel product shot dropped onto a slide and dragged to a corner is still a 4000-pixel file inside the ZIP. A 200-pixel thumbnail dropped onto a full-bleed placeholder is still 200 pixels when it comes out. The extractor does not know what you meant. It knows what is stored.

That is why recovered assets often look larger and sharper than the deck suggests — and why they sometimes look worse. Screenshotting the slide captures display size, compression, and whatever sits under the picture: a theme colour, a crop, a title that is not in the image. Unpacking ppt/media/ skips the layout. You get the bytes that were embedded, including crops PowerPoint did not apply to the stored file.

The tool will not rasterize a slide. DrawingML — themes, masters, placeholder inheritance, fonts — is a layout engine this path does not run. Faking a screenshot of the slide would look like the deck without being the deck. The job is media. It is not a render.

Limits sit on the page before you pick a file: 200 MB and 300 slides. A .ppt is the old binary format. It is not a ZIP, and there is no ppt/media/ to open. Re-save as .pptx from PowerPoint or Keynote.

Slides built only from text, shapes, and theme colours have nothing in ppt/media/. The listing will be empty. That is not a failure. Those objects were never embedded files.

Names follow add-time, not slide number

PowerPoint discards the filename you dropped. The stored names are usually image1.png, image2.jpeg, media2.mp4, and so on. The extractor keeps those names. It does not recover hero-pack-shot-final.psd. That string is gone.

Order follows when each asset was added to the package, not which slide it sits on. image1 is not “slide 1.” Slide 12 can own image3 because that picture was dropped first. Identifying files means looking at thumbnails, not sorting by slide number and hoping the integers match.

The on-page grid shows 60 assets. The rest is in the ZIP download. Do not assume the grid is the full set. If the deck embedded 80 pictures, the ZIP is the delivery, not the first page of thumbnails.

Video and audio in ppt/media/ come out the same way: original container, original bitrate, original duration. A clip scaled to a tiny player on the slide is still the file that was inserted. Trimming on the slide does not trim the stored media. You may get more seconds than the animation showed. That is the embedded original, not a bug.

PPTX to text is the sibling for words. Different ZIP parts (ppt/slides/, notes). It will not give you the pictures. The media extractor will not give you the speaker notes. Pull both if you need both. Neither job typesets a new slide.

Extraction is a ZIP listing, not a new picture

A .pptx is an Open Packaging Convention archive. The tab reads it with the File API, lists ppt/media/, and offers the parts back as downloads. Nothing is uploaded. No account. Network off, it still unpacks. The free tools shelf is this class of work: local, no model, no credit meter.

/cost is the map of generate meters — per second, per thousand characters, per megapixel, per job. Unpacking a PNG that is already in the deck is not on that map. Versely has no free generation tier. This tool is free because it never calls a model.

The failure mode is sending a screenshot of the slide, or the whole deck, to an image model with “extract the photos.” You will get new pictures. They will be the model’s guess at a stock photo that looked a bit like your product shot. You will have spent credits. You will still not have the original file.

Same mistake, video flavour: asking a video model to “recreate the slide animation.” You wanted media2.mp4 out of the package. You paid for a hallucinated motion graphic at display size.

If the slide only shows a picture of text — a screenshot, a chart exported as pixels — that asset is an image in ppt/media/. The extractor will give you that image. It will not OCR it. Words that live only as pixels stay pixels.

Strip EXIF later; generate even later

The embedded original can still carry a sidecar. A JPEG dropped from a phone often has GPS in APP1. Extraction does not strip it. It returns the file as stored.

Image metadata stripper is the next local job when the recovered still should leave without coordinates. It reads the JPEG EXIF — including GPS — then re-encodes through canvas so the download has none of it. PNG text chunks are listed the same way. That is a new bitmap, not a surgical edit of the original bytes. Canvas JPEG is a new compress; export PNG if you need a lossless pixel copy. Nothing is uploaded. The original on disk is untouched.

Do not launder a strip through a model. Generating “a copy of this photo without GPS” spends credits on a cousin of the still. Labels drift. Faces drift. The canvas already drops APP1.

After you have the original — and after you have stripped it, if that is the brief — decide whether you still need a generate. Animate a recovered pack shot on image-to-video, billed on that row, shown before you confirm. Clean a still on an image row. Caption a clip that was already in the deck with an editor pass, not a generate of the clip.

Order is the whole product: unpack the original at drop-in resolution; identify by thumbnail because the names are image1 and the order is add-time; strip EXIF if the sidecar is the leak; generate only if you still lack pixels. Reverse that order and you will pay for a picture of a picture at slide size.

FAQ

Will the images match the size they appear on the slide?

No. You get the embedded originals. Display size on the slide is a scale. The file in ppt/media/ is the source. A tiny on-slide crop of a large photo still extracts large. A full-bleed 200-pixel drop-in still extracts at 200 pixels.

Why are the filenames image1.png and media2.mp4?

PowerPoint renamed them when they were stored and discarded the originals. Order roughly follows add-time, not slide number. Identify by looking at the thumbnails. Do not assume image3 lives on slide 3.

Can it read an old .ppt?

No. Re-save as .pptx. The older format is not an Open Packaging ZIP, so there is no ppt/media/ to open. Limits on the .pptx path are 200 MB and 300 slides.

Should I send the extracted JPEG to a model to “clean GPS and sharpen” in one step?

Strip GPS locally with the image metadata stripper. Sharpening or a new crop is a later generate, billed in credits, on the file you already recovered. Mixing those jobs is how an original becomes a cousin.