Image · Open Source · Commercial Use · ComfyUI · 4K Texture
Microsoft's 4B-parameter image-to-3D model producing up to 1536³ resolution PBR textured assets. Uses novel O-Voxel representation that handles arbitrary topology including open surfaces and non-manifold geometry. Generates high-fidelity assets in 3-60 seconds depending on resolution. Fully open source under MIT license.
Follow TRELLIS (Microsoft) on X (Twitter) for the latest news, updates, and announcements.
Follow @MSFTResearch
Microsoft's 4B-parameter image-to-3D model producing up to 1536³ resolution PBR textured assets. Uses novel O-Voxel representation that handles arbitrary topology including open surfaces and non-manifold geometry. Generates high-fidelity assets in 3-60 seconds depending on resolution. Fully open source under MIT license.
28% win rate · 646 votes across modes
Community rating · out of 5
Community rating
6 ratings · 2 written reviews
Rank, ELO and votes are frozen season results. Cost, runtime and sample outputs are version metadata, not season snapshots.
Arena Rank
#12 / 23
Textured ELO
1011
Geometry ELO
1023
Win Rate
51%
1,060 wins · 1,006 losses
Total: 2,066 votes
Average Generation Cost
$0.42
Typical runtime: 8.0 min
Nearby ranks in Textured
#11 · 1012 ELO
In December 25, TRELLIS.2 ranked #12 (1011 ELO), compared with #11 (1012 ELO) for Hi3D v2.0.
Hi3D v2.0 holds a 1 point advantage, so TRELLIS.2 fans tend to focus on prompts that highlight its strengths.
During this season, TRELLIS.2 won 51% of its judged matches in this arena, compared with 51% for Hi3D v2.0.
Hi3D v2.0 often returns meshes quicker (7.0 min) while TRELLIS.2 focuses on fidelity.
#13 · 1007 ELO
In December 25, TRELLIS.2 ranked #12 (1011 ELO), compared with #13 (1007 ELO) for Rodin Gen 1.5.
TRELLIS.2 maintains a 4 point lead, which usually translates to steadier community votes.
During this season, TRELLIS.2 won 51% of its judged matches in this arena, compared with 48% for Rodin Gen 1.5.
Rodin Gen 1.5 often returns meshes quicker (2.5 min) while TRELLIS.2 focuses on fidelity.
#10 · 1017 ELO
In December 25, TRELLIS.2 ranked #12 (1011 ELO), compared with #10 (1017 ELO) for Tripo AI v3.0.
Tripo AI v3.0 holds a 6 point advantage, so TRELLIS.2 fans tend to focus on prompts that highlight its strengths.
During this season, TRELLIS.2 won 51% of its judged matches in this arena, compared with 56% for Tripo AI v3.0.
Tripo AI v3.0 often returns meshes quicker (4.0 min) while TRELLIS.2 focuses on fidelity.
#14 · 1001 ELO
In December 25, TRELLIS.2 ranked #12 (1011 ELO), compared with #14 (1001 ELO) for Hunyuan 3D v2.5.
TRELLIS.2 maintains a 10 point lead, which usually translates to steadier community votes.
During this season, TRELLIS.2 won 51% of its judged matches in this arena, compared with 52% for Hunyuan 3D v2.5.
Hunyuan 3D v2.5 often returns meshes quicker (3.0 min) while TRELLIS.2 focuses on fidelity.
What was the Top 3D AI rank for TRELLIS.2 in December 25?
In December 25, TRELLIS.2 finished at #12 with 1011 ELO and 2,066 community votes.
How much does it cost to generate with TRELLIS.2?
TRELLIS.2 reports an average generation cost of $0.42 per test with a typical runtime of 8.0 min.
How did TRELLIS.2 compare with Hi3D v2.0 in December 25?
In December 25, TRELLIS.2 ranked #12 (1011 ELO), compared with #11 (1012 ELO) for Hi3D v2.0. Hi3D v2.0 holds a 1 point advantage, so TRELLIS.2 fans tend to focus on prompts that highlight its strengths. During this season, TRELLIS.2 won 51% of its judged matches in this arena, compared with 51% for Hi3D v2.0. Hi3D v2.0 often returns meshes quicker (7.0 min) while TRELLIS.2 focuses on fidelity.
