Invisible Text Extractor

It looks for the nine zero-width characters it recognises (the Invisible Text Hider writes eight of them) — U+200C, U+200D, U+2060, U+200E, U+200F, U+200B, U+00AD, U+2062 and U+2064 — counts them, and tries to read them as a stream of bits: one bit per character (two characters), two bits (four) or three bits (eight), grouped into bytes. The first four bytes must be a length that fits in the data; if they do, that many bytes are the payload. Other invisible characters, such as U+FEFF, are neither counted nor removed.

Auto-Detect tries six fixed character-set combinations and keeps the best one. Ranking: a result that passed AES-GCM authentication comes first; a reading whose length field accounts for every byte ranks above one that does not; then readable text ranks above encrypted-looking data, which ranks above noise; a reading with damage beyond repair loses points. Positions do not matter and no seed is needed, because the decoder reads every matching character in order. A message made with a character set outside those six needs Manual mode with the matching mode.

'Hidden Message Extracted' means the bytes decode as readable UTF-8. 'Encrypted Payload Found' means the payload matches the decoder's encrypted-format checks (the length field of this site's encrypted format) or is random-looking enough (entropy above 7.4 bits per byte) to warrant further inspection. Neither alone proves it is encrypted: compressed or random data looks the same. If you expect an AES-GCM message, enter the password and decode again. A wrong password and a missing password give the same result — the tool cannot tell them apart, and it never shows a green 'decrypted' mark unless AES-GCM authentication passed.

Short payloads are not judged by entropy alone. Entropy estimates are less reliable for short payloads, because a small sample may not represent the underlying data well, so for short payloads the extractor also checks the structure of this site's encrypted format. Neither high entropy nor a matching structure alone proves that data is encrypted: compressed or random data looks the same.

The score is computed from the character counts only: 30 for any zero-width character, up to 30 for density above 0.1% or 1%, 25 for two or more types, and 15 when exactly two types occur in similar numbers. Legitimate text can reach a moderate score: scripts such as Persian and the Indic languages use U+200C and U+200D in ordinary spelling, and emoji sequences use U+200D. A high score means 'worth decoding', not 'a message is hidden'; a low score does not prove there is none.

Work down this list. (1) Is it the original text? Any application that strips or replaces invisible characters changes the stream and breaks it — a single removed character shifts every bit after it. (2) Is a password needed? Encrypted data shows the 'Encrypted Payload Found' result. (3) Try Manual mode with each mode in turn. (4) The characters may be innocent typography. (5) For other hiding places in text, try the Text Steganography Extractor or the Whitespace Steganography Extractor.