tools
Labelbox Model Foundry
Labelbox Model Foundry uses foundation models to automate data labeling, enrichment, and model-assisted workflows.
Model Foundry lets AI and machine-learning teams select foundation models, generate predictions, pre-label datasets, enrich data, and send predictions to human reviewers in Labelbox Annotate. It supports computer-vision and natural-language tasks such as object detection, image classification, segmentation, text generation, translation, summarization, and named-entity recognition.
Foundry is used through Labelbox Catalog, Annotate, and Model. It does not replace human review for quality assurance. Usage is billed monthly through model inference costs and Labelbox units; the exact model costs are shown in the product.
Features
- Generate model predictions from selected data rows
- Pre-label datasets with foundation models
- Enrich datasets with model-generated metadata
- Support object detection, classification, and image segmentation
- Support text generation, translation, summarization, and question answering
- Configure ontologies and confidence thresholds
- Preview model runs before processing full datasets
- Send predictions to Annotate for human review
Use cases
- Pre-label images for object detection and classification
- Enrich datasets with generated text or metadata
- Review model predictions with human annotators
- Compare model outputs before selecting a production workflow
- Extract named entities from documents and text
- Export predictions for offline analysis or downstream pipelines