Tencent-Hunyuan/InstantCharacter ? reverse-engineered prompt

Build me a simple web app for InstantCharacter that lets someone upload one reference image of a person or character, type a prompt like what they want that character doing, and generate a new image that keeps the same identity. I’d like it to work with the existing model pipeline, and have a clean interface where I can also choose a style if I want, like a Ghibli or Makoto Shinkai look. Please include a normal run mode and a lower memory mode for people with less VRAM, so it still works on a wider range of GPUs.

Make the app easy to launch locally, with a clear input and output flow, a seed option for repeatable results, and sensible defaults so it feels good right away. If anything needs current docs or model details, look them up online and use the safest approach. Also add a simple demo script so I can run the generation from the command line if I do not want the web UI.