Feed-forward system that turns an RGB image or video into editable, textured per-object 3D meshes in under a minute.
Updated EU status checked Last shown
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
Fire3D (FIRE3D: Feed-forward Interactive 3D Scene Reconstruction) is a research framework from the University of Illinois Urbana-Champaign and Cornell University. From an unsegmented RGB image or casual video, it detects the objects in a scene, estimates their 6-DoF poses and oriented boxes, and reconstructs each one as a complete textured mesh with PBR appearance. A static background instance holds structural surfaces.
The pipeline uses Pi3X to estimate point maps and camera poses, lifts DINOv3 features into a shared 3D representation, and then completes each entity with structure, shape and material flows. It builds on the TRELLIS.2 SC-VAE and adds HC-VAE, a second hierarchical compression stage (32× smaller latent) that lets up to 16 objects be processed in parallel on one GPU. Generation uses a fixed 12-step flow-sampling schedule and no test-time optimization. The authors report up to twelve textured instances in under a minute, including preprocessing.
The outputs are amodally complete, independently editable assets intended for games and interactive applications. According to the project page, it reports leading detection accuracy on perception benchmarks and competitive rendering quality, compared with systems such as ShapeR and SAM 3D Objects. A YouTube review states that code and training code have been released, but no repository link was in the provided material.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Reconstructs simulation-ready 3D scenes from photos or video, with each object as its own mesh with materials. Runs in under a minute and handles 16 objects in parallel on one GPU; code and training code released.
Closest neighbours by meaning, not just by category.
Model
Feed-forward 3D scene reconstruction model that outputs triangle primitives directly usable as meshes for simulation.
Model
Research model that decomposes a single image or scene mesh into object-level 3D meshes with consistent spatial layout.
Model
Feed-forward 3D Gaussian Splatting model that reconstructs scenes from arbitrary input views, with or without camera poses.
Research
Post-training method that refines feed-forward 3D reconstruction models (Pi3, DA3, VGGT) on unlabeled video.
Research · ByteDance
ByteDance research pipeline that turns indoor captures into editable, simulation-ready 3D scenes of separate object meshes.
Model
Research framework that turns posed images with instance masks and 3D boxes into per-object meshes for cluttered scenes.
