Trellis.2

Adoption verdict: Trellis.2 looks worth tracking and piloting, but the current evidence supports “high-potential image-to-3D draft tool” more than “proven production-grade replacement.” The strongest evidence here is fal’s official X announcement of the Trellis.2 LoRA Trainer plus a small number of demo posts, which supports real momentum around the model and surrounding workflow tooling. At the same time, the Zhihu sources are mostly paper explanations and result breakdowns rather than extensive independent benchmarking, so attention proof is stronger than usability proof.

In practice, this is not a general 3D modeling app like Blender and not an engineering CAD tool. A more accurate analogy is an image-to-3D asset generator or a 3D draft machine. The value discussed in the evidence is speed: starting from one reference image, getting a textured or stylized 3D result that can be refined later. With fal’s 3D LoRA training, there also appears to be room for style or asset-domain adaptation.