What problem does it solve? Converting a single 2D image into a usable 3D asset with shape and PBR textures is error-prone: bad input images produce melted geometry, VRAM limits cause out-of-memory failures, and misconfigured shape or paint parameters waste hours of GPU time. This Skill codifies proven recipes for running Tencent Hunyuan3D 2.1 inside ComfyUI so each generation succeeds on the first queue. ## Core Features & Use Cases - Input image preparation rules: guidance on centered single objects, transparent backgrounds via rembg, and silhouette quality that determines most of the output quality. - VRAM recipes for 12GB GPUs: concrete parameter sets for shape (~10GB) and texture painting (low_vram_mode, texture_size 1024, view_size 512) on cards like the RTX 4070, plus OOM troubleshooting steps. - Shape and PBR parameter tables: steps, guidance_scale, octree_resolution, max_num_view, and texture_size with recommended ranges for preview versus production quality. - Export and QA workflows: GLB/OBJ/STL export guidance, Blender-based headless QA rendering, and a CPU-only vertex-color fallback when full PBR is not viable. - Use Case: A game artist uploads a product photo of a prop, follows the checklist to generate a textured GLB on a 12GB GPU, then validates the mesh in Blender before importing it into the game engine. ## Quick Start Use the hunyuan-3d-v21 skill to turn my input image 'prop.png' into a textured 3D model for a game, running on a 12GB GPU.
