Dev trains Krea 2 LoRA plus tiny image-to-3D model for cheap voxel generation
Angaisb_ · x · 2026-11-14
Developer Angaisb shared an experiment for generating cheap voxel 3D models: a Krea 2 image-gen LoRA trained on renders of a few high-quality voxel models from two angles, followed by a tiny from-scratch model that converts those images into real 3D models.
The results aren't good yet, but the author thinks this image-gen-to-3D pipeline could become interesting as future image models get smarter.
More from Multimodal
- Open-source Blender add-on runs NVIDIA Kimodo, ARDY and MotionBricks locally for AI motion generation — bdsqlsz · 2026-11-14
- NeurIPS 2025 best paper: diffusion models start memorizing at ~300 steps per image — Mastbubbles · 2026-11-14
- Sci-fi film KILL/SAVE team releases "fake ads from 2045" made with AI video — killsavecom · 2026-11-14
- Stella & Shango Part 2: Viking-themed AI video short continues — stellaXaiXvip · 2026-11-14
- Fable 5.2 trick: one recolored 3D character populates every spectator NPC — chongdashu · 2026-11-14
- Making Midjourney obey prompts: low --stylize plus --raw, in v8.2 — michaelrabone · 2026-11-14