Note Note License: source-available, Creative Commons license strictly for non-commercial use Training data: 12k+ fully licensed images, balanced benchmark Provenance: Strong: fully licensed, consented data; but not open Environmental impact: Very low: lightweight segmentation model (published) Energy disclosure: Not disclosed

Note File: .obj, .glb | vertex color -> you can use InstantTexture to generate UVmap License: MIT (code + weights) Training data: Curated Objaverse subset, CC-BY Provenance: Strong: curated and licensed subset, disclosed Environmental impact: Very low: mesh in <0.5s on one GPU (published) Energy disclosure: Not disclosed; training cost published: 5 days, 176 A100 GPUs

Note File: .glb | Quality: very low License: MIT this is not an AI model! It is a tool to convert vertex-colored .obj to textured .glb

Note File: .glb | UVmap License: MIT (code, weights, dataset) Training data: TRELLIS-500K (Objaverse-XL, ABO, 3D-FUTURE, HSSD, Toys4k) Provenance: Good: curation disclosed, but Objaverse lineage is mixed Environmental impact: High: 1.2B-4B parameters (published) Energy disclosure: Not disclosed

Note File: .glb | UVmap ! Runs on ZERO GPU (try it at home, not during the workshop) License: MIT (code, weights, dataset) Training data: TRELLIS-500K (Objaverse-XL, ABO, 3D-FUTURE, HSSD, Toys4k) Provenance: Good: curation disclosed, but Objaverse lineage is mixed Environmental impact: High: 1.2B-4B parameters (published) Energy disclosure: Not disclosed

Note File: .glb | UVmap ! Runs on ZERO GPU (try it at home, not during the workshop) License: Apache 2.0 (code + own weights); depends on Zero123++ weights under CC-BY-NC 4.0 (non-commercial; outputs free) Training data: Not documented Provenance: Mixed: open code and free outputs, but non-commercial dependency and undocumented training data Environmental impact: Moderate: 3D assets within 10 seconds (published); Energy disclosure: Not disclosed;

Note File: .glb | UVmap ! Runs on ZERO GPU (try it at home, not during the workshop) License: MIT (code + weights) Training data: Human scans (THuman2.1, CustomHumans, CAPE) + SD2.1 lineage Provenance: Weak: human-scan consent issues, biometric fit concerns Environmental impact: High: over 40GB VRAM, multi-view diffusion (published) Energy disclosure: Not disclosed

Note File: .glb | UVmap ! Runs on ZERO GPU (try it at home, not during the workshop) License: MIT (code only) Training data: Undocumented, no training code released Provenance: Weak: no provenance at all Environmental impact: Moderate: about 30s per mesh (published) Energy disclosure: Not disclosed