Slide Image Sizer

Somewhere in most decks there is a face that is slightly too wide. Nobody decided to widen it. The photograph was one shape, the slide was another, and a piece of software settled the difference by pulling the picture until its edges met the edges of the slide. The audience sees the result before it reads a word.

The shape in question is the aspect ratio, width divided by height. A slide that is 16 units wide for every 9 tall has a ratio of about 1.78. One that is 4 wide for every 3 tall has 1.33. Put a 1.33 picture on a 1.78 slide and something has to give.

The shapes a slide can be

PowerPoint’s shapes are well recorded. The default slide has been 16:9 since PowerPoint 2013, and the size table that the LibreOffice project keeps for trading files with it gives the dimensions in inches: PowerPoint Widescreen is 13.333 by 7.5 inches, and On-screen Show (4:3), the older standard shape, is 10 by 7.5 inches. There is a second 16:9 option, On-screen Show (16:9), at 10 by 5.625 inches, and a 16:10 one at 10 by 6.25 inches. The two 16:9 sizes have the same aspect ratio, because 13.333 divided by 7.5 and 10 divided by 5.625 both come to 1.78, so they look the same on screen. The larger one simply has more surface.

Google Slides offers the same family under Page setup: Standard (4:3), Widescreen (16:9), Widescreen (16:10) and Custom, with the custom size typed in inches, centimeters, points or pixels. Its help page prints no dimensions for the presets, which is a quiet reminder that the number that matters is the ratio.

Inches are a strange unit for something that will only ever be light on a wall. A slide has no pixels of its own. Inside the file it is a page measured in English Metric Units, and there are 914,400 of them in one inch and 360,000 in a centimeter, a unit created so that both systems divide evenly. A slide 7.5 inches tall is 6,858,000 units tall. The standard lets a side run from 914,400 units to 51,206,400, which is 1 inch to 56. The pixels come later, from whatever shows the deck.

How many pixels the room can use

A high-definition picture, as the broadcasting standard defines it, is 1920 by 1080 samples at 16:9. Ultra-high definition doubles each side to 3840 by 2160, with a further step at 7680 by 4320. A projector or a display built to one of those formats cannot show more than it has. A full-slide picture of 1920 by 1080 pixels gives a high-definition projector one image pixel for every pixel it can light.

The arithmetic changes when the slide and the display disagree. A 4:3 slide on a 16:9 high-definition display is limited by height, so it occupies 1440 by 1080 pixels, with unused screen to either side. Anything larger is carried in the file and discarded at the wall.

Presentation software has an opinion as well, and the file format leaves room for it. The standard gives every presentation an attribute named autoCompressPictures, which says whether the program should automatically compress all the pictures in the deck, and the schema treats it as switched on unless a file says otherwise. The standard’s overview ties the feature to the spread of digital cameras and scanners and to the wish for small files. It names no resolution, so how far a picture is reduced is each program’s own decision. A 3840 pixel picture sent to fill a slide may not stay 3840 pixels wide once a program has saved the deck.

Too few pixels is the worse error, because nothing can put back what was never recorded. Stretch an 800 by 600 picture across a 1920 by 1080 slide and each of its pixels must cover 2.4 pixels of slide in each direction. The software can only guess at what lies between them. The picture gets bigger and softer at the same rate.

Three ways to settle it

There are only three ways to put one rectangle on another, and the style language of the web names them more exactly than most slide software does. Cover: the picture keeps its aspect ratio while filling the entire box, and whatever hangs over the edge is cut off. Contain: it keeps its aspect ratio while fitting within the box, and the leftover area shows whatever lies behind. Fill, in that vocabulary, means the picture is simply sized to the box, ratio or no ratio. The tool on this page uses plainer words for the same three: Fill for cover, Fit for contain, Stretch for the last.

Numbers make the trade plain, and the sample photograph in the tool will do. It is 2400 by 1600 pixels, a 3:2 shape, and the slide is 1920 by 1080. To cover the slide the photograph is scaled by the larger of 1920/2400 and 1080/1600, which is 0.8. At that scale the slide shows only 1350 of its 1600 rows, so 250 go, 125 from the top and 125 from the bottom when the crop is centered. To fit instead, the smaller factor applies, 0.675, and the photograph lands as 1620 by 1080 with a bar of 150 pixels on each side. Stretching uses both factors at once and makes everything 19 percent too wide.

Cropping, then, costs the sample about a sixth of its rows. Which sixth is the real decision.

Left alone, the crop takes equal amounts from both ends, which suits a centered subject and beheads an off-center one. The remedy is a focal point: mark the place that must survive, and let the crop window slide until that place sits in its middle. The same idea exists in web layout, where a position property determines how a picture is aligned inside its box. A focal point has one hard limit. Near an edge of the photograph the window stops, because there are no pixels beyond the edge to show.

Fitting keeps every pixel and pays in empty slide. The usual filler is black, and on a projected image black bars read as a smaller picture on a screen that promised a larger one. A gentler filler is the photograph itself: enlarged until it covers the slide, blurred until no detail survives, darkened a little, and laid behind the sharp copy.

The blur is the Gaussian kind that graphics standards describe, a weighted average that spreads each pixel across its neighbors. The same standard notes that three successive box blurs approximate it to within roughly 3 percent, which is how the tool computes it. The result has the picture’s own colors and none of its content, so the eye takes it for surroundings and not for a second image.

It is not always the right filler. A blurred screenshot or a blurred chart is only a smear, and a logo on a blurred copy of itself looks like a printing fault. Diagrams want a flat color, preferably one the deck already uses.

Stretching is the option to refuse. It is harmless only when the two shapes already agree, in which case it is not stretching at all. Otherwise circles become ovals, and a face is the last thing that survives being widened.

What the file has to carry

Format decides size. PNG is lossless by design: its specification describes a format for the lossless, well-compressed storage of raster images, which is what a screenshot or a diagram with hard edges needs. JPEG is the most widely used lossy format for still images and is particularly useful for photographs, while lossy compression of content that needs sharpness, such as diagrams or charts, can give unsatisfactory results. Photographs go to JPG. Drawings and text go to PNG.

The difference is not small. Saved from the tool in Chrome at 1920 by 1080, the sample photograph is about 1.1 megabytes as PNG and about 80 kilobytes as JPG at 90 percent quality, roughly a thirteenth of the size. Thirty such slides make a deck of about 32 megabytes one way and about 2.4 the other.

The deck itself is less mysterious than its icon suggests. The .pptx format has been PowerPoint’s default since PowerPoint 2007, and its content is marked up in two XML vocabularies, PresentationML and DrawingML, standardized as ECMA-376 and as ISO/IEC 29500. The container is a ZIP archive, a format PKWARE has published since 1989, and the only compression its rules allow is DEFLATE, a lossless method. The Open Packaging Conventions arrange what is inside as a collection of files, called parts, and the relationships among them. Rename a copy of a deck from .pptx to .zip and any unzip program will list them.

What is inside is a small bureaucracy. A minimum presentation, in the Open XML documentation Microsoft publishes, has a presentation part, a slide master, a slide layout, a theme and a slide, and the standard adds a presentation properties part to the list. Every slide is a separate XML file. A picture slide adds one thing more: the image file, stored in the package and reached from the slide through a relationship. The slide’s XML says where the picture goes and how large it is, in the same units as the slide. For a picture from edge to edge on a widescreen slide those numbers are 0, 0 and 12,192,000 by 6,858,000.

That is all a deck built by the tool contains: one picture per slide, already cropped or fitted, placed to the edges.

What sizing cannot do

Sizing a picture in advance fixes it to one slide shape. Change the deck from 16:9 to 4:3 afterward and every picture has to be made again, because the crop and the bars are now part of the image. It cannot sharpen a small photograph, and it cannot bring back the rows a crop removed. A deck of pictures is also only that. It has no titles, no speaker notes and no animation until someone adds them.

So choose the shape of the slide first. Then, for each picture, decide what you can afford to lose: its edges, part of the slide, or the truth of its proportions. Only the first two are worth choosing.