LightCyan01/TrainKit — reverse-engineered prompt

Build me a Windows desktop app called TrainKit that makes it easy to prep image datasets for AI training. I want a clean modern interface with a sidebar and three main tools, caption images, upscale images, and batch rename files. Each screen should let me pick an input folder, output folder, and model path where needed, then start processing and watch progress and logs in the app.

For captioning, use local image caption models with customizable prompts so I can generate captions for a whole folder. For upscaling, support common local upscale models and make it GPU accelerated. For renaming, let me rename and organize lots of image files with flexible naming patterns. The backend should start automatically when the app launches so it feels like one simple desktop tool.

Please make it match the current README and get the project into a working state from source, including build and run flow for Windows. If anything is unclear, check the latest docs online and fill in the gaps sensibly.