Practical overview
When it fits
Use PassGAN when a research or assessment workflow needs generated password candidates or model training from an authorized password dataset.
Who it is for
- Security researchers studying password guessing.
- Practitioners conducting authorized password-strength assessments.
- Machine-learning users exploring GAN-based password generation with curated datasets.
Documented uses
- Generate password candidates from the pretrained model for research or assessment.
- Train a password-generation model from a dataset.
- Compare generated candidates with held-out data as part of a separate research workflow.
Requirements & limits
The repository is public and released under the MIT License.
- The documented setup requires CUDA 8 and the Python requirements installed from the repository.
- The README warns that training on datasets of 100 MB or more may take several hours on a GTX 1080.
- The author characterizes the reported reproduction as not exhaustively analyzed and questions how it compares with prior RNN and Markov approaches.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Generate candidates with the pretrained model
Use the documented sampling command to write password candidates to a local file.
Before you start
- CUDA 8 pre-installed
- Python dependencies installed with pip install -r requirements.txt
- The repository's pretrained model and checkpoint
Steps
Install the repository requirements after CUDA 8 is available.
Run sample.py with the pretrained input directory and checkpoint.
Set the output file, batch size and number of samples; the README example writes one million samples to gen_passwords.txt.
Expected outcome
A local text file containing generated password candidates.
Guide limitations
- The documented command samples candidates; it does not itself provide a password-strength score.
- The README's reported test-set result is described as not exhaustively analyzed.
Train a model on a password dataset
Use train.py with a dataset file to produce model training output.
Before you start
- CUDA 8 pre-installed
- Python dependencies installed with pip install -r requirements.txt
- A password dataset file available as training data
Steps
Prepare an authorized dataset file.
Run train.py with --output-dir and --training-data arguments.
Expected outcome
Training output and checkpoints saved in the chosen output directory; the README example uses default hyperparameters and 200,000 iterations, saving checkpoints every 5,000.
Guide limitations
- The README says training on a large dataset of 100 MB or more can take several hours on a GTX 1080.
- The guide does not establish performance for other datasets or hardware.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 15 October 2026
- Original website title
- brannondorsey/PassGAN · Official homepage title · retained in its source language