Video frame extraction guides, organised by task

Nearly every request falls into one of four shapes. Sampling means taking evenly spaced frames across a stretch of video — one per second, one every ten — to review, annotate, or feed into something else. Targeting means finding one specific frame and saving it at full quality. Building means the frames are raw material for something else: a GIF, a thumbnail, a contact sheet. Preparing means the frames become a dataset, where consistency and provenance matter more than any individual image.

The settings that follow from each are genuinely different. Sampling is decided by an interval and lives or dies on getting the frame count right before you start. Targeting depends on knowing the source frame rate so that stepping moves exactly one frame. Building usually needs fewer frames than people expect and a deliberate choice of output format. Preparing needs naming and splitting decisions made before annotation begins, because they are expensive to change afterwards.

The guides below are grouped that way rather than by tool.

The first is that every guide has to be useful whether or not you use anything here. That means real menu paths, real commands, and real parameter values for the other tools — the exact FFmpeg filter string, the full path to VLC's scene filter, the dialog options in Photoshop's Video Frames to Layers. If a guide only made sense as an argument for switching tools, it would not be worth publishing.

The second is that each one says where this site is the wrong choice. FFmpeg remains the right answer for anything scripted, batched, or running on a server, and for codecs a browser will not decode. VLC is better when the source is a network stream or when you want an unattended capture across a long recording. A full editor is correct when the still needs retouching, colour management, or has to match a graded sequence. Browser-based extraction covers a narrower case: one-off jobs, no installation, and files that should not leave the machine.

That last point comes up more than you might expect. Client footage under contract, recordings containing identifiable people, coaching video of minors, pre-publication research material — in all of these the constraint is not a preference but someone else's policy, and it decides the tool before any feature comparison begins. The guides on confidential footage and on dataset preparation both cover how to verify that a tool really processes locally, which takes about a minute and replaces a promise with a fact.

Finally, none of these tools obtain permission for you. They read files you already have. Whether you may keep, publish, or share the frames you extract depends on the licence of the original video and, where people are recognisable, on the rules that apply to recordings of them. That question is worth settling before the extraction, not after.