Digital 3D objects used in games and animation are often required to be compositional; that is, decomposed into semantically meaningful parts. Recent 3D generation methods can produce high-quality compositional objects conditioned on image or text prompts. Yet, such global conditioning lacks the precise part-level controllability required for professional creative workflows. To address this, we introduce MultiCube, a novel compositional 3D generation method that provides explicit, independent control over both the semantics and spatial arrangement of each part. MultiCube takes as input a global text prompt, a text schema specifying the desired parts, and a spatial layout indicating the bounding boxes of the parts in the given schema. It outputs a 3D object composed of distinct meshes, one per specified part, that adhere to the given semantic and spatial conditions. Our approach employs a two-stage diffusion process, first generating a schema- and layout-aligned monolithic mesh, then decomposing the mesh into individual parts simultaneously. A novel Part Layout Adapter is used to encode per-part conditions independently of the other parts. Experiments demonstrate that our method can generate high-quality compositional 3D objects with precise part-level control, including those with unique layouts difficult to achieve with text or image prompting alone. Project page: https://multi-cube.github.io
Figures & tables
Figure 2: Part-level semantic control. Users can specify different part semantics for the same spatial layout, resulting in unique output designs.
Figure 3: Text- or 2D layout-conditioned image generation models, and the image-conditioned 3D generation methods that rely on their output, struggle to follow prompts with complex part layouts. With 3D layout conditioning, MultiCube can generate shapes aligned with precise part layouts.
Figure 4: MultiCube uses a two-stage diffusion-based architecture. In Stage 1, it synthesizes a monolithic object conditioned on the input text prompt and part layout (a). In Stage 2, it takes the object latents produced in Stage 1 and decomposes them into multiple shapes, one for each specified part, guided by the part labels and bounding boxes (b). Spatial conditions are encoded with a Part Layout Adapter in Stage 1 and a lightweight embedding in Stage 2.
Figure 5: Part Layout Adapter. The adapter encodes each (part label, bounding box) condition pair into a set of part latents. The label is encoded using Qwen-VL and the bounding box goes through a Fourier embedding before being transformed into a latent sequence using a Q-Former [ 20 ] .
Figure 6: Qualitative ablation results. We compare our full model with the following ablated setups: a naive part condition encoding with no adapter, an entangled part condition encoding, and using semantic conditions only during multi-part generation.
Figure 7: Ablation study. We run head-to-head comparisons of our model against ablated setups on part-level alignment and geometric quality.
Figure 8: Qualitative multi-part generation comparison. Our method conditions on an input shape, spatial layout, and semantic part labels. CubePart [ 55 ] and the “PatchAlign3D [ 12 ] + HoloPart [ 45 ] ” pipelines condition on shape + semantic part labels. FullPart [ 9 ] and OmniPart [ 47 ] are image- and spatial layout-conditioned; OmniPart additionally requires 2D segmentation and 3D voxel inputs. Compared with baselines, our method exhibits cleaner part decompositions and more detailed geometry.
PartObjaverse-Tiny
Part-Level
Holistic-Level
CD ↓
F-score ↑
Box IoU ↑
CD ↓
F-score ↑
PartCrafter
0.612
0.346
0.176
0.142
0.781
PartPacker
0.325
0.594
0.321
0.084
0.925
GT seg/vox + OmniPart
0.058
0.940
0.753
0.038
0.991
FullPart
0.118
0.828
0.723
0.105
0.871
PA3D + HoloPart
0.228
0.698
0.405
0.048
0.971
Table 1: Quantitative evaluation for multi-part generation. We compute Chamfer distance and F-scores at part and holistic levels, alongside bounding box alignment at the part level. Our method strongly improves part-level accuracy. Note that OmniPart is given a ground-truth 2D segmentation and 3D voxels as input.
Figure 9: Full pipeline generation results. MultiCube generates compositional 3D objects from input text prompts and part layouts. Here, the part bounding box layouts are automatically generated using GPT-5.1 [ 36 ] ; hence, the generation process from prompt to output shape is fully automatic.
MultiCube win rate
End-to-end
Box IoU ↑
VLM Sem.
VLM Qual.
GT render + FullPart
0.723
82.9%
89.0%
GT render/seg + OmniPart
0.343
84.9%
82.2%
MultiCube (end-to-end)
0.717
—
—
Table 2: End-to-end evaluation on PartObjaverse-Tiny. Box IoU measures compliance with the input part layout. The VLM columns report MultiCube’s win rate against each baseline on part semantic alignment and geometric quality, with ties counted as half wins. Both baselines additionally receive a ground-truth rendering; OmniPart also receives a ground-truth 2D segmentation. MultiCube is conditioned on text and layout alone.
Figure 10: Additional generation results, using part layouts from PartObjaverse-Tiny.
Figure 11: Generalization to scenes. Though our model is trained mostly on object-level shapes, it generalizes to producing layout-conditioned scenes. Users can control the position, size, and type of each object in the scene.
Figure 12: Failure cases. Our model may produce strange outputs if the input bounding boxes are wildly incorrect (left). Additionally, it occasionally produces overlapping parts (right).
Appendix figures & tables2 assets
Supplementary material from the paper’s appendix.
Appendix
Figure 14
Method
Input Part Labels
3D Inputs
2D Inputs
Monolithic Shape
Monolithic Voxels
Part Boxes
Rendered Image
Part Seg
Grip, Body, Magazines, Barrel, Tail
PartCrafter
✓
PartPacker
✓
GT seg/voxels + OmniPart
✓
✓
✓
✓
FullPart
✓
✓
Appendix
Table 3: Inputs to multi-part generation methods. Inputs include ground-truth part labels, shapes, voxelized shapes, part bounding boxes, rendered 2D images, and 2D segmentations from PartObjaverse-Tiny.
Interactive 3D assets used in games and simulation are typically decomposed into specific semantic parts to support animation, physics, and scripted behaviors, yet most generative 3D models produce either monolithic meshes or arbitrary part decompositions that cannot be aligned with application-specific requirements. We present CubePart, a generative framework for open-vocabulary, part-controllable 3D mesh generation that exposes part structure as an explicit inference-time control signal. Given a global text prompt and a user-defined parts schema expressed as an open-ended list of part names, our method generates a set of meshes - one per schema element - that assemble into a coherent object while respecting the specified semantic structure. To enable this capability, we introduce a scalable data pipeline to construct a large open-vocabulary, part-labeled 3D dataset, along with a two-stage generative architecture that separates global shape synthesis from part-level decoding. We demonstrate that the resulting assets can be directly integrated into game engines and driven by animation and behavior scripts without manual post-processing. Project Page: https://cubepart.github.io/
Understanding and generating 3D objects as compositions of meaningful parts is fundamental to human perception and reasoning. However, most text-to-3D methods overlook the semantic and functional structure of parts. While recent part-aware approaches introduce decomposition, they remain largely geometry-focused, lacking semantic grounding and failing to model how parts align with textual descriptions or their inter-part relations. We propose DreamPartGen, a framework for semantically grounded, part-aware text-to-3D generation. DreamPartGen introduces Duplex Part Latents (DPLs) that jointly model each part's geometry and appearance, and Relational Semantic Latents (RSLs) that capture inter-part dependencies derived from language. A synchronized co-denoising process enforces mutual geometric and semantic consistency, enabling coherent, interpretable, and text-aligned 3D synthesis. Across multiple benchmarks, DreamPartGen delivers state-of-the-art performance in geometric fidelity and text-shape alignment.
Part-level 3D assets are essential for editing, reassembly, and interaction, yet recovering such structure from a single image remains challenging due to occlusion, ambiguous boundaries, and the need for coherent multi-part reasoning. Existing approaches struggle to achieve both controllable part-level generation and coherent multi-part structure, as part identity and spatial allocation are typically inferred implicitly. We present Seg3DParts, a segmentation-grounded framework for controllable part-level 3D generation from a single image. By treating segmentation as an explicit grounding signal, our method defines part identity during generation, enabling each component to be anchored to a corresponding image region. To ensure coherent assemblies, we introduce structured cross-part interaction that allows components to exchange global context throughout the generative process. As a result, Seg3DParts directly generates well-aligned part meshes in a shared canonical space without post-hoc alignment, supporting flexible and controllable decomposition. We further introduce PartObjectNet, a large-scale dataset with over 200K objects and 1M annotated parts. Experiments demonstrate that Seg3DParts achieves superior geometry quality, cross-part coherence, and part-level controllability over existing methods.
Jiantao Lin, Meixi Chen, Yingjie Xu +5
The Hong Kong University of Science and Technology (Guangzhou) · Knowin AI · The Hong Kong University of Science and Technology