How it works
Files are assembled byte by byte in your browser from each format's own specification, then padded using a mechanism that format actually permits, so the result is both the exact size you asked for and genuinely valid in a real reader.
Every format has somewhere it may carry bytes no reader interprets, and the padding goes there: a JUNK chunk in RIFF containers such as WAV and AVI, a PADDING metadata block in FLAC, an ancillary chunk in PNG, an entry's extra field in ZIP-based formats like DOCX and XLSX, whole sectors in ISO 9660, and comment lines in text formats such as JSON, CSV and YAML.
Structure costs bytes before your content exists, so each format has a minimum size and the size list starts there rather than offering a file that could not exist. Combining content options raises that floor, because each option adds a section of its own.
Any format can also be generated broken, in three specific ways: invalid (structural or data type errors), corrupted (broken or unreadable headers) and incomplete (truncated part-way through). The damage is applied to the bytes rather than the file name, so parsers reject the file for the reason you would expect.
Because generation is local, the values you choose and the files you produce are never transmitted: there is no server to receive them, no account to store them against and no upload step. The site does use Google Analytics and Google AdSense, which set cookies in your browser, but neither ever receives the contents of a generated file.