Show HN: Tenderness – open-source synthetic data generation for VLM/OCR (github.com)

Tenderness is an innovative open-source library designed for synthetic data generation specifically for vision-language models (VLM) and optical character recognition (OCR). Unlike traditional methods that involve post-processing noisy datasets through OCR and heuristics, Tenderness directly renders text into structured document formats—such as images, SVGs, and PDFs—ensuring precise control over layout and eliminating the need for inference. This approach allows for the generation of large-scale synthetic datasets that provide rich structural supervision, enabling more accurate model training and validation. Significantly, Tenderness empowers developers to create consistent, reproducible document layouts from predefined primitives such as text blocks, images, and tables. Its minimal flexbox layout engine enhances positioning automation, while its capabilities extend to extracting detailed bounding boxes and offering rich typography options. By removing manual annotation requirements, Tenderness streamlines dataset creation, which has profound implications for the AI/ML community in advancing layout understanding systems and establishing benchmarks for performance evaluation. Developers can easily integrate the library into their workflows with its user-friendly pipeline setup, making it a valuable tool for enhancing the quality and reliability of document-based AI projects.