Base64 Encoder & Decoder

Encode UTF-8 without tripping btoa

window.btoa is a Latin-1 function. Pass it a string with emoji or a non-BMP character and the browser throws. Data URLs, HTTP basic auth, and JSON web token segments all expect Base64 (or base64url) of bytes, not of UTF-16 code units. This page walks TextEncoder to get UTF-8 bytes, builds a binary string, then calls btoa. Decode reverses with atob and TextDecoder. Conversion is live: there is no submit button.

Mode Encode treats the left pane as plaintext. Mode Decode treats it as Base64. Standard alphabet uses + and / with = padding. URL-safe uses - and _ and strips padding, which is what JWTs and many query parameters need. Decode always accepts both alphabets.

bytes = TextEncoder.encode(str)
btoa(binaryFromBytes)
inflation % = (encoded - raw) / raw x 100

Telemetry reports raw UTF-8 bytes, encoded string bytes, inflation percent, and line count of the current input. Empty input zeros the cards and clears output. Swap exchanges the panes and flips Encode/Decode so you can round-trip a sample. Load sample text includes café, naïve, and a wave emoji so you can confirm Unicode does not throw.

Corrupt input and clipboard

Garbage characters in decode mode are caught. The alert region is assertive. The script does not stop. Copy uses the Clipboard API with a 1px on-screen textarea fallback if permission is denied. Labels return to Copy output after two seconds (Copied! or Copy failed).

This is a text tool. It will not hash files, wrap MIME 76-character lines, or implement Base32. For JSON pretty-print after you decode a payload, open the JSON Formatter and Minifier. For compact tokens that look like Base64 but are three segments, use the JWT Decoder instead of decoding the whole string as one blob.

Padding: Standard output includes = when needed. URL-safe output does not. Some servers reject unpadded strings; some reject padded URL-safe strings. Match the consumer. Copy the output you see, not a remembered variant from another site.

When +33.3% is the expected result

Three bytes become four Base64 characters. That is +33.3% before padding. Short strings pad up to a multiple of 4, so inflation can sit above 33%. Decode mode still reports raw as the recovered plaintext bytes and encoded as the Base64 you typed, so the ratio stays meaningful. If inflation is 0% you are probably looking at empty input.

Nothing is posted. Treat this as a scratch pad for snippets, not a vault. Clear wipes both panes. If you need YAML rather than Base64, use the JSON to YAML Converter on the same site.

Frequently Asked Questions (FAQ)