Benchmarking text-to-3D generative AI outputs for additive manufacturing

Abstract

This study presents a single-instance benchmark of five commercial text-to-3D artificial intelligence tools—Meshy, PrintPal, Sloyd, 3D AI Studio and Tripo—for generating engineering components intended for fused filament fabrication (FFF). Three reference CAD models were used: an L-bracket, a print-in-place hinge and a cylindrical mounting bracket. Standardized prompts specified dimensions, functional features and manufacturability constraints. One generated output and one printed specimen were evaluated for each tool–prompt condition (n = 1); therefore, the findings describe the tested instances. The assessment combined mesh and pre-print checks, scaling and repair records, dimensional measurements for the L-bracket, point-cloud and deviation-map comparisons, criterion-fulfilment scoring for the hinge and cylindrical bracket, and repair-burden assessment. The tools inferred broad object categories but did not reliably preserve engineering constraints. Among the tested outputs, Sloyd showed the closest L-bracket correspondence and the highest hinge criterion score, whereas PrintPal and 3D AI Studio showed the highest cylindrical-bracket criterion scores. No output produced a functional print-in-place hinge or a coherent cylindrical mounting fixture. Direct text-to-3D tools are therefore useful for ideation, but still require CAD reconstruction, validation and process-aware preparation before engineering use in FFF.