Developer tinyBigGAMES has made available Gemma4.pas, a native Object Pascal implementation of Google’s Gemma 4 multimodal models, bringing local text and vision AI directly to Delphi applications.
Rather than relying on external inference frameworks, Gemma4.pas demonstrates how modern multimodal LLMs can be integrated into native Delphi applications with support for conversational AI, image understanding, and streaming responses. The project serves as both a reference implementation and a foundation for developers building local AI experiences in Object Pascal.
Gemma4.pas includes support for text generation, image analysis, streaming inference, GGUF model loading, and interactive chat, making it possible to build applications that can both understand images and carry on natural conversations using locally hosted Gemma 4 models.
As part of the tinyBigGAMES ecosystem of native AI projects, Gemma4.pas showcases how state-of-the-art multimodal models can be brought to Delphi without leaving the Object Pascal ecosystem.
Explore how to run Google’s Gemma 4 vision models from native Delphi applications.
