Teaching Local Models to Generate 3D Assets via Contour-First Principles

On November 12, 2026, an author writing in the AI section of DEV Community published a walkthrough on training local models for 3D generation using the open source agent Flash and the local model Onyx. The piece uses a 3D chair as its running example: ask a model for a 3D chair and you get five boxes, a box for the seat, a box for the back, and four thin boxes for legs, roughly in the right places. Turn the camera and one leg is floating a centimetre above the floor, the back is sunk into the seat, and the whole thing is the size of a house because nobody said what unit "1" was. The author's diagnosis is that the model is not failing at geometry; it simply has never been taught how people who make 3D models actually work. The update's main contribution is to teach the silhouette-first rule of 3D modeling together with image-guided techniques, meaning the model is trained to reason from shape outlines and reference images before committing to detailed geometry. Beyond this article, no follow-up updates or independent reporting have appeared, so the current state of the work is the published walkthrough itself.