Sensei-HQ

Katachi turns a photo of something you love into pieces you can cut on a desktop laser or a Cricut. Snap or upload a picture; Katachi finds the items and the layers, cleans them up, lays out sheets for your machine and shows how they go together. You can say what you want, typed or spoken, or tap a button: the chat and the buttons run the same actions, with one undo. Three crafts show the range: a layered cut (a paper cat, built to the stack I described: a solid white base, black with the whiskers cut through, red, white flowers, gold, each layer keeping the outline), a collage (28 earrings from one tray photo, paired by shape and evened out), and a set (five overlapping bookmarks shot at an angle, squared up, rebuilt and made one size). Each costs 2–3 cents of Meta model calls. The rule that keeps it precise: models see and decide; code measures. Muse Spark and SAM 3.1 perceive and plan; the model never emits a coordinate, and every line that gets cut is computed in code. It is a working prototype, honest about its limits (on /story). Built during the hackathon; pre-existing: my open-source UI library Rokkit and the usual open-source stack. See it: https://katachi.sensei-hq.com/story (the film, the three examples with their photos and steps, the slides). Try it: invite code : KBRX-Y36G-6UJK at https://katachi.sensei-hq.com (tick the three boxes)