How a PNG goes into a PDF without loss
A PNG stores each row of pixels after a small prediction step, and then compresses the result with deflate. A PDF has a filter, FlateDecode with PNG predictors, that reads exactly that. So for most PNGs the compressed data can be copied into the PDF as it is. Nothing is decoded and nothing is rounded, which is why the pixels in the PDF are the pixels in the file.
This works for grey PNGs, palette PNGs and colour PNGs with 8 or 16 bits per channel, as long as they are not interlaced. Palette PNGs stay palette pictures and grey PNGs stay grey, so a flat graphic does not grow. A colour profile inside the PNG is carried into the PDF with the picture.
Some PNGs cannot be copied directly: those with an alpha channel or a transparent colour, interlaced ones, animated ones and ones that record a rotation. They are decoded and stored again, still without loss, so the pixels are the same, but the PDF can be bigger or smaller than the PNG file. Only the first picture of an animated PNG is used. Text notes and other metadata in the PNG are not copied.
Transparency: what the PDF does with it
A PNG can have clear and half clear pixels. A PDF has no such thing as a transparent picture on its own, so ClipLime stores the colour of the picture and its clear parts as two separate pictures: the colours, and a grey picture used as a soft mask. The page itself is filled with the page colour, and a PDF viewer lets that colour show through wherever the mask says the picture is clear. Soft edges and shadows keep their softness. Opaque pixels are stored exactly. Half clear pixels look the same over any colour to within one step out of 255, a rounding that comes from the way browsers hold transparent pictures, and the colour values hidden under fully clear pixels are not kept.
Pick the page colour to suit where the PDF will be used: white for printing, or a dark colour for a dark presentation. The swatches cover white, light grey, grey, black and cream, and the colour box takes any other.
If a picture has only a few clear pixels, fewer than one in a thousand, such as a screenshot with rounded window corners, it is flattened onto the page colour instead. That gives a simpler, smaller file that looks the same.
Screenshots and graphics
A screenshot of 1920 by 1080 pixels with the page size set to Same as the image and 200 DPI becomes a page of 9.6 by 5.4 inches, about 244 by 137 mm, and each screen pixel stays one picture pixel. The same screenshot on a portrait A4 page is drawn at about 232 DPI and has white bands above and below it. Both are sharp, but the page that follows the picture wastes no space.
A small picture on a big page is enlarged, and enlarging cannot add detail. If a result looks soft, the picture is too small for the page. The tool warns when a picture would print below 100 DPI. Use a smaller page, the Same as the image size, or a bigger source picture.
A very large screenshot, such as a 4K capture of 3840 pixels across, is more than a page can show. With Smaller file it is reduced to the chosen DPI and kept lossless, which keeps text readable but loses fine detail. Choose Original quality if you need every pixel.
PNG photos and big PDFs
PNG is lossless, and the noise in a photograph does not compress well without loss. The same photo as a PNG is typically several times bigger than a JPG, and a PDF built from it is as big as the PNG. That is not a fault of the PDF. It is the price of keeping every pixel.
Smaller file looks at the pixels of each PNG. If the picture is a photograph, it is reduced to the chosen resolution and saved as JPG at quality 84. If it is flat, such as a screenshot, a logo or a chart, or it has transparency, it stays lossless. A picture is never made bigger: if saving it again would not make it smaller, the PNG data goes in as it is. The details after building tell you what happened to each picture.