Comparative assessment of AI-based tools for generating 3D models from text-to-3D and image-to-3D workflows

AI tools for 3D content generation are expanding digital modeling for additive manufacturing, but their suitability for engineering-oriented Fused Filament Fabrication (FFF) remains limited when target geometry requires dimensional control, valid meshes, feature orientation, component separation, and functional interpretation. This study evaluates the first-response behavior of ChatGPT, Gemini, and DeepSeek using four benchmark conditions: a solid cube, a cube with a central through-hole, a monolithic compliant tweezer, and a male/female snap-fit pair. The protocol reproduced a common user interaction: one standardized request, the first returned output, and no iterative regeneration or prompt refinement. Manually modeled and printed standards were used as targets. Outputs were evaluated as upstream FFF candidates by STL/G-code integrity, bounding-box correspondence, watertightness, winding consistency, component count, feature preservation, snap-fit plausibility, and repair burden. ChatGPT produced the highest number of accepted first-response outputs across the tested conditions; Gemini performed well for simple geometries but fragmented or simplified more demanding cases; and DeepSeek showed feature, orientation, and topology failures. Image input improved scale and layout communication but did not replace mesh inspection or engineering-feature verification. AI-generated STL files should be treated as preliminary FFF design candidates, not direct engineering-grade components.