NIST Model Card Generator

NIST has not published an official model card template. A model card is a short document describing an AI model's intended use, data, performance, limitations and risks, an idea introduced by Google researchers in 2019. This tool structures one using themes from the NIST AI Risk Management Framework (AI RMF 1.0, NIST AI 100-1), which encourages transparency and documentation across its Govern, Map, Measure and Manage functions.

No. The NIST AI RMF is a voluntary framework, and NIST does not certify or approve model cards. This tool is not an official NIST product and is not endorsed by NIST. It uses the framework's transparency and risk themes as a structure. Whether your documentation meets any requirement depends on what it says and on the rules that apply to you.

Model cards support AI governance, vendor and procurement reviews, internal risk assessments, and documentation needs under regulations such as the EU AI Act. They help users and auditors understand what a model is for and where it can fail. A model card on its own does not make a system compliant with any law or standard.

No. The tool runs entirely in your browser and nothing is uploaded. Form contents and saved versions are kept only in the page's memory, so they disappear when you reload or close the tab. When you download a card, the file is saved to your device, so download before leaving the page.

Templates fill the form with starting text for an LLM, a vision model or a classifier. Anything you need to supply, such as metrics, datasets and safeguards, is an [ADD: ...] placeholder. Replace every placeholder with your own information. Cards exported while placeholders or required fields remain are marked as drafts.

You can still export, but the card is marked DRAFT and lists what is missing, so an incomplete card is not mistaken for a finished one. Required fields are model name, primary use cases, limitations, fairness and bias metrics, and risks.

Include measured results from validation and testing, and name the dataset and test conditions: accuracy, F1-score, precision and recall, AUC-ROC, robustness under adversarial conditions, fairness metrics such as demographic parity or equal opportunity, and any other relevant indicators.

Yes. Markdown files can be edited in any text editor, such as VS Code, Obsidian or Notepad++. JSON exports can be modified programmatically or with any JSON editor. Within the tool you can save up to five snapshots while the page stays open.

Markdown (.md) is human-readable and suits documentation, GitHub and wikis. JSON (.json) is machine-readable and suits model registries and automated systems. In the JSON file, fields you left empty are null rather than filler text.

ML engineers, data scientists, AI product managers, compliance and risk teams, and anyone deploying AI models who needs to document them for stakeholders, auditors or regulators.