FMXExpress has released ONNX Image Tagger, a demonstration of how modern AI image classification and tagging models can be integrated directly into Delphi applications using native ONNX Runtime inference.
While the sample application automatically generates tags for images, the real story is the underlying TONNXRuntime wrapper by hshatti, which brings Microsoft’s ONNX Runtime to Delphi and FreePascal developers through a native Object Pascal API.
The project demonstrates how Delphi applications can run AI models locally without Python, cloud APIs, or external AI services.
The demo showcases:
- Local AI image tagging
- ONNX model inference
- Native Delphi integration
- Offline AI workflows
- Computer vision applications
- High-performance inference pipelines
Because ONNX Runtime supports models from a wide range of AI ecosystems, the same foundation can be used for image classification, object detection, OCR, embeddings, vision-language models, and many other machine learning workloads.
An interesting detail for Windows developers is that ONNX Runtime is already included with modern versions of Windows through the Windows AI platform, while the latest standalone runtime packages and execution providers are available directly from Microsoft. This makes ONNX one of the easiest ways to bring local AI capabilities into Delphi applications.
Although ONNX Image Tagger is a practical example, it also serves as a blueprint for a much larger idea: using TONNXRuntime as the foundation for deploying local AI models across Delphi applications on Windows, Linux, and macOS.
