InternGPT

Why we included this project

InternGPT lets you steer a large language model with your mouse instead of typing every instruction. For vision-heavy work, where you often need to refer to specific parts of an image, this makes the chat interface noticeably more precise. The repo wires together several foundation models: SAM for segmentation, DragGAN for drag-based editing, ImageBind for cross-modal understanding, and Husky, a fine-tuned vision-language model that powers the multimodal dialogue. Researchers and developers who want to see how these pieces fit into one interactive demo will find a lot to reuse here. Keep in mind the project is still under construction, so treat it as a reference implementation and starting point rather than production software.