A video frame extraction method performed by a computing device includes: obtaining a key frame timestamp table of a video based on a first sampling stride, the first sampling stride being adjusted according to a length of the object video; obtaining a timestamp of a key frame to be extracted from the key frame timestamp table based on a second sampling stride, the second sampling stride being determined according to a quantity of key frame timestamps in the key frame timestamp table and a quantity of to-be-extracted key frames; reading, according to the timestamp, frame data of the key frame; and generating, based on a hard decoding result of the frame data, a thumbnail corresponding to the key frame.