Frame

A frame is a single still image from a sequence of images. A video, after all, is not real motion but a fast succession of many photos. If they are shown quickly enough one after another, they merge in the eye into smooth motion. How many individual images are shown per second is called the frame rate, abbreviated fps for “frames per second.” Cinema classically runs at 24 frames per second, games often at 60 or more. The German word for it would simply be Einzelbild (single image), but in technical jargon almost everyone says frame.

Why programs think in individual images

Computers cannot process motion directly. They need something countable, and that is precisely what frames are. A ten-minute video at 30 frames per second consists of 18,000 individual images. Each of these is, for the machine, initially its own separate image that must be stored, transmitted, or analyzed.

This immediately creates a cost problem. If an AI is supposed to understand a video, it must in principle look at every single one of these images. That is many times more demanding than assessing a single photo. That’s why many systems resort to a trick: they only look at every tenth or every fiftieth image. For a soccer match, this is often enough, but for a fast hand movement, it can cause exactly the decisive moment to be lost.

With games, too, the number of frames is a hard limit. An image must be fully rendered before it appears on the monitor. If the graphics card only manages 20 images per second, the game visibly stutters. That’s why anyone talking about graphics performance is almost always talking about frames per second.

From raw image to compressed image sequence

A single frame is, technically, a grid of pixels. In Full HD, that’s around two million dots, each with its own color value. Uncompressed, one minute of video would take up so many gigabytes that streaming would be impossible. Video formats solve this with a simple observation: two consecutive images are almost always very similar to each other.

That’s why only occasionally is a complete image stored, a so-called keyframe. All the images in between store merely the differences from the previous one. If, in a scene, only a person’s mouth is moving, that amounts to very little data. This is exactly why a video sometimes looks briefly garbled on a poor connection: a keyframe is missing, and the differences are applied to the wrong base image.

With AI video generators, the task is reversed. They must produce frames that match one another. A single beautiful image is easy, but the person must not suddenly be wearing a different shirt in the next image. This image-to-image stability is called temporal consistency and is currently the hardest problem in this field.

Frames in streams, games, and industry news

In everyday life, you encounter this term wherever moving images are playing. Video platforms often offer a choice of 30 or 60 frames per second. Phone cameras advertise slow motion at 240 frames per second: you capture a very large number of frames and play them back slowly. When stepping forward frame by frame in a player, you jump from frame to frame.

In tech news, the term usually comes up in connection with graphics cards. Techniques such as Frame Generation let the hardware calculate additional in-between images that don’t actually exist in the original. A neural network estimates what the image between two real frames should look like. The result appears smoother but can show artifacts during fast motion.

A common misconception: more frames do not automatically mean a better picture. A 60-hertz monitor can only display 60 images per second; anything beyond that is wasted. And a movie at 24 frames is by no means worse-looking, just familiar. The right frame rate depends on what is being shown, not on which number is bigger.