🎨 NeMo Data Designer
Generate high-quality synthetic datasets from scratch or using your own seed data.
📄 Read the technical report — Architecture, programming model, and case studies for multimodal synthetic data generation.
Welcome!
Data Designer helps you create synthetic datasets that go beyond simple LLM prompting. Whether you need diverse statistical distributions, meaningful correlations between fields, or validated high-quality outputs, Data Designer provides a flexible framework for building production-grade synthetic data.
What can you do with Data Designer?
- Generate diverse text, structured data, and images using samplers, LLMs, image models, or existing seed datasets
- Build multimodal workflows with image, audio, and video context
- Connect generation to external tools through local or remote MCP servers and capture interaction traces
- Control relationships between fields with dependency-aware generation
- Validate and score outputs with Python, SQL, custom validators, and LLM judges
- Extend Data Designer with plugins for specialized data sources, generation capabilities, and processing steps
- Preview, resume, and monitor generation from small experiments to large runs
Quick Start
1. Install
pip install data-designer
[!NOTE]
To apply maintained minimum dependency versions for known vulnerabilities, use the optional security constraints file:
pip install data-designer -c "https://raw.githubusercontent.com/NVIDIA-NeMo/DataDesigner/main/constraints.txt"
With uv, use uv pip install or uv add with the same -c URL. See dependency security for details.
[!NOTE]
This project will download and install additional third-party open source
software projects. Review the license terms of these open source projects
before use.
Or install from source:
git clone https://github.com/NVIDIA-NeMo/DataDesigner.git
cd DataDesigner
make install
2. Set your API key
Start with one of our default model providers:
Grab your API key(s) using the above links and set one or more of the following environment variables:
export NVIDIA_API_KEY="your-api-key-here"
export OPENAI_API_KEY="your-openai-api-key-here"
export OPENROUTER_API_KEY="your-openrouter-api-key-here"
3. Start generating data!
import data_designer.config as dd
from data_designer.interface import DataDesigner
# Initialize with default settings
data_designer = DataDesigner()
config_builder = dd.DataDesignerConfigBuilder()
# Add a product category
config_builder.add_column(
dd.SamplerColumnConfig(
name="product_category",
sampler_type=dd.SamplerType.CATEGORY,
params=dd.CategorySamplerParams(
values=["Electronics", "Clothing", "Home & Kitchen", "Books"],
),
)
)
# Generate personalized customer reviews
config_builder.add_column(
dd.LLMTextColumnConfig(
name="review",
model_alias="nvidia-text",
prompt="Write a brief product review for a {{ product_category }} item you recently purchased.",
)
)
# Preview your dataset
preview = data_designer.preview(config_builder=config_builder)
preview.display_sample_record()
Do more with Data Designer plugins
Plugins extend the data sources, generation capabilities, and processing steps available in your Data Designer workflows. The DataDesignerPlugins repository contains optional NVIDIA-maintained packages that add these capabilities.
| What you want to do | Plugin package |
|---|
| Use files from GitHub or local Git repositories as seed data | GitHub seed reader |
| Generate question-answer data for retriever training, with document chunking, embedding-based deduplication, and export to Automodel-compatible training formats | Retrieval SDG toolkit |
Data Designer provides the generation framework. Plugin packages add specialized capabilities to your existing workflow and can also provide ready-made workflows with their own CLI.
After installing Data Designer, browse and install compatible packages from the default NVIDIA catalog:
data-designer plugin list
data-designer plugin info github
data-designer plugin install github
data-designer plugin installed
The NVIDIA catalog is configured by default. Install plugin packages in the same Python environment as Data Designer. Their extensions are discovered automatically after installation; follow each package’s usage guide to configure them in your workflow.
Browse available plugins · Discovery and installation guide · Build your own plugin
What's next?
📚 Learn more
📝 Documentation
Data Designer documentation is available at docs.nvidia.com/nemo/datadesigner.
Contributors should edit docs prose under fern/. Tutorial notebook source remains in docs/notebook_source/*.py; generated notebooks and Fern artifacts are not the source of truth. The legacy MkDocs archive remains available on GitHub Pages for releases 0.5.7 and older.
🔧 Configure models via CLI
data-designer config providers # Configure model providers
data-designer config models # Set up your model configurations
data-designer config list # View current settings
🤖 Agent Skill
Data Designer has a skill for coding agents. Just describe the dataset you want, and your agent handles schema design, validation, and generation. The skill is tested with Claude Code and Codex.
Install via skills.sh:
npx skills add NVIDIA-NeMo/DataDesigner
After installation, invoke the data-designer skill or describe the dataset you want and the skill will kick in.
🤝 Get involved
This repository supports agent-assisted development — see CONTRIBUTING.md for the recommended workflow.
- Contributing Guide – How to contribute, including issue triage before external PRs and agent-assisted workflows
- GitHub Issues – Report bugs or make a feature request
Telemetry
Data Designer collects telemetry to help us improve the library for developers. This data is not used to track any individual user behavior. It is used to see an aggregation of which models are the most popular for SDG. We will share this usage data with the community.
Disable with NEMO_TELEMETRY_ENABLED=false. More details →
Top models (YTD)
Aggregate model usage across synthetic data generation jobs, year-to-date 1/1/2026–10/27/2026:
Last updated on October 27, 2026
License
Apache License 2.0 – see LICENSE for details.
Citation
If you use NeMo Data Designer in your research, please cite our technical report:
@misc{greco2026nemodatadesignerextensible,
title = {NeMo Data Designer: An Extensible Framework for Multimodal Synthetic Data Generation},
author = {Johnny Greco and Nabin Mulepati and Andre Manoel and Eric Tramel and Kirit Thadaka and Mike Knepper and Dhruv Nathawani and Dane Corneil and Yev Meyer and Alex Watson and Maarten Van Segbroeck},
year = {2026},
eprint = {2609.17699},
archivePrefix = {arXiv},
primaryClass = {cs.AI},
url = {https://arxiv.org/abs/2609.17699},
}
Telemetry & privacy
NeMo Data Designer includes an optional function to share anonymous telemetry data with NVIDIA for product improvement. Data collected is limited to names of models used and token counts (input and output). No user or device information is collected. This data is used to prioritize product improvements and will be shared in aggregate with the community. It is not used to track any individual user behavior.
You may opt out of telemetry collection at any time. Opting out applies only to data collection by the NeMo Data Designer library itself.
Use of third-party endpoints, including NVIDIA Build: NeMo Data Designer can be configured to use various inference endpoints, including build.nvidia.com (NVIDIA Build). If you choose to use NVIDIA Build or any other third-party endpoint, that endpoint's own terms of service and privacy practices apply independently of this library. Any opt-out you exercise within NeMo Data Designer does not extend to data collection by your chosen endpoint. NVIDIA Build is intended for evaluation and testing purposes only and may not be used in production environments. Do not submit any confidential information or personal data when using NVIDIA Build.
