image-tryon

image-tryon is a virtual try-on skill for the Starchild platform that lets users preview how an item looks on a person — clothing, accessories, hairstyles, makeup, glasses, hats, shoes, and watches. It solves the problem of visualizing purchases or style changes without a physical fitting, and always operates on two images: a person photo and a garment/item photo.

Rather than reimplementing proxy or billing plumbing, the skill calls a provided Python script (try_on) that reads both local files, base64-encodes them, and sends them to the fal.ai image service as data URIs, or accepts public HTTPS URLs instead. It exposes parameters for category (which selects an intent-appropriate default prompt), an optional custom prompt override, model choice (a fast NanoPro default around 25 seconds or a higher-quality GPT Image 2 option around 150 seconds), and output aspect ratio. It documents careful result delivery: never hand the user the raw fal.media URL (which is served with restrictive CSP headers), and instead use the already-downloaded local file path, embedding it inline on web or sending it via Telegram/WeChat helpers. Detailed intent-recognition and photo-requirement tables map natural-language requests to the right category.

It targets Starchild agent users and end users wanting quick visual try-on across fashion and beauty categories. Note that the underlying script files were not available for review, so security assessment relies on the documented behavior.