What you lose when you compress a photo
Why 50 KB looks fine for one picture and terrible for another.
JPEG compression works by throwing away information the eye is bad at noticing. How much it can throw away before you notice depends entirely on what is in the picture.
Where the detail goes
The image is divided into small blocks, and each block is described by how much it varies at different levels of fineness. Compression keeps the coarse description and discards the fine one. A smooth sky is almost entirely coarse, so it survives brutal compression. Hair, grass, text and foliage are almost entirely fine detail, so they turn into mush.
The artefacts to look for
- Blocking: faint squares across smooth areas such as skin and sky.
- Ringing: a shimmer along hard edges, worst around text.
- Colour bleed: colour spreading past the edge it belongs to, because colour is stored at half resolution.
Pixels versus quality
When a file will not fit a limit, you can lower the quality or reduce the pixels. Below roughly quality 50 the artefacts become obvious; at that point reducing the dimensions gives a far better-looking result for the same file size. A sharp 600-pixel image beats a mangled 1200-pixel one every time.
Generation loss
Every time a JPEG is opened, edited and saved, it is compressed again, and the damage accumulates. Edit from the original each time rather than from a file you compressed yesterday, and keep that original somewhere safe.
When not to use JPEG
Screenshots, scans of text, logos and line drawings should be PNG. JPEG smears the sharp edges that make text readable, and for images with few colours PNG is often smaller as well as better.
Re-encoding through a browser also strips EXIF metadata, including the GPS coordinates a phone records. That is usually something to be glad about before sending a photo to a stranger.