How to use the Password Generator
Last verified 2026-08-08 · about 6 minutes to read
The short version
- 1.Set the length. Twenty characters is a good default for anything stored in a password manager. Length matters far more than which symbols you include.
- 2.Generate. Press Generate for a fresh batch. Nothing is stored, logged or transmitted.
- 3.Save it in a password manager. A generated password is only useful if you never have to remember it. Copy it straight into a manager.
Why this is worth getting right
Two implementation details separate a real password generator from a demo. First, crypto.getRandomValues rather than Math.random, whose state is recoverable from a handful of outputs. Second, rejection sampling instead of a modulo: taking a random 32-bit integer mod 72 makes the first 40 characters of the alphabet measurably more likely than the rest.
Method 1 — use the Password Generator on this site
The fastest route. It runs entirely in your browser, so nothing is uploaded and there is no queue to wait in. No account is needed and there is no daily limit.
- Set the length. Twenty characters is a good default for anything stored in a password manager. Length matters far more than which symbols you include.
- Generate. Press Generate for a fresh batch. Nothing is stored, logged or transmitted.
- Save it in a password manager. A generated password is only useful if you never have to remember it. Copy it straight into a manager.
Method 2 — do it without this site
Worth knowing, because a tool you cannot replace is a dependency rather than a convenience. Most tasks in the security division have a command-line or built-in equivalent; it is usually more setup and less convenient, but it works offline and it is scriptable, which matters once you are doing something a hundred times instead of once.
What this tool will not do
Every tool has an edge. These are ours for password generator, stated up front so you find out here rather than halfway through a deadline:
- The entropy figure describes the generator, not a password you chose yourself. It says nothing about a human-picked password.
- Crack times assume an offline attack on a weak hash. Against bcrypt or Argon2 they are far longer.
- "Require one of each type" slightly reduces entropy; the number shown is the unconstrained figure.
Questions people ask
Are these passwords generated on a server?
No. They are generated in your browser by crypto.getRandomValues. Nothing is sent over the network, nothing is logged, and closing the tab destroys them. A password generator that round-trips to a server has already failed at its one job.
How long should a password be?
Sixteen characters or more for anything that matters, and twenty if a password manager is doing the remembering. Length is the dominant factor: adding one character to a 72-character alphabet multiplies the search space by 72, while adding a symbol type to a fixed length does far less.
What does entropy in bits actually mean?
It is the base-2 logarithm of the number of passwords the generator could have produced. Every extra bit doubles that number. Under 50 bits is weak against an offline attack; 80 bits or more is comfortable; 128 bits is beyond any foreseeable brute force.
How is the crack time estimated?
It assumes one trillion guesses per second — an offline attack with GPUs against a weakly hashed password file. That is a pessimistic assumption on purpose. An online attack against a rate-limited login is millions of times slower.
Why does "require one of each type" reduce entropy?
Because it removes every password that lacks one of the types, shrinking the possible set. The reduction is fractions of a bit at normal lengths, and worth accepting when a site enforces composition rules. The bits shown here are the honest figure for a purely random draw.
Is Math.random good enough for this?
No. Math.random uses a fast non-cryptographic PRNG whose internal state can be recovered from a small number of observed outputs, after which every future value is predictable. This tool uses crypto.getRandomValues with rejection sampling to eliminate modulo bias.