Luce: Relightable Gaussians for 3D Asset Generation

Mayank Singh; Michele Stoppa; Alvise Memo; Rui Yu; Sree Harsha Kalli; Srimanth Gunturi; Muhammad Ahmed Riaz; Behrooz Shahsavari

cs.CVAug 25, 2026

Luce: Relightable Gaussians for 3D Asset Generation

Authors: Mayank Singh, Michele Stoppa, Alvise Memo, Rui Yu, Sree Harsha Kalli, Srimanth Gunturi, Muhammad Ahmed Riaz, Behrooz Shahsavari, +2 more

Organizations: Apple

Abstract

High-fidelity image-to-3D generation requires a 3D representation that captures both geometry and appearance. However, preserving fine detail across the physically based rendering (PBR) modalities needed for relighting remains challenging. To address this, we propose Luce, a 3D representation that unifies geometry and PBR materials within a voxelized multimodal Gaussian cloud, using dedicated Gaussian primitives for albedo, metallic-roughness, and surface normals. A variational autoencoder compresses this representation into a unified material-aware latent space. A rectified-flow transformer generates this latent from a single image using multi-layer features from a pretrained image encoder that preserve both semantic context and fine spatial detail. The latent is then decoded into relightable PBR Gaussians and an optional textured mesh with a tangent-space normal map. On Toys4K, Luce achieves state-of-the-art single-image-to-3D generation, improving FID by 28% over the strongest baseline. We further evaluate Luce on a benchmark of AI-generated images depicting diverse subjects and materials, where it improves the CLIP image-alignment score over the best baseline (0.8519 vs. 0.8299). Luce generates relightable, geometrically accurate, and materially faithful assets that preserve fine details such as text, logos, and inscriptions.

Figures & tables

Panel of Figure 1 — Figure 1: Our method Luce generates relightable 3D assets from a single image. Each asset consists of Gaussians encoding three modalities of a physically based rendering (PBR) material: albedo, metallic-roughness, and surface normals. These modalities enable standard PBR shading, rendering the Gaussians under novel illumination. From left to right, the columns show: the input condition image; the generated PBR modalities stacked vertically—albedo (top), metallic-roughness (middle, metallic in Red, roughness in Green channel), and surface normals (bottom); and the generated 3D asset rendered under three environment maps, previewed as small strips above each render (top row).

Panel of Figure 2 — Figure 2: Legible text on generated 3D assets. Luce compared with LiTo ( Chang et al., 2026 ) , TRELLIS 2 ( Xiang et al., 2026 ) , and TRELLIS ( Xiang et al., 2025 ) on single-image-to-3D generation; each row shows the input condition image and one rendered view of each method’s generated 3D asset. LiTo generates assets aligned to the input view, whereas other methods generate in a canonical orientation. Luce keeps surface text and markings legible where baselines distort them.

Figure 3: Overview of Luce. (Top) Representation and SLatVAE. Given a 3D PBR asset, here a Toys4K ( Stojanov et al., 2021 ) sample, we render multiview images and fit per-modality Gaussian splats (albedo, metallic-roughness, normals) on a sparse voxel grid. A structured latent VAE (SLatVAE) encodes this representation into a compact, diffusible latent and decodes it back to PBR Gaussians. (Bottom) Generation pipeline. Given a single input image, a sparse-structure flow first predicts the sparse voxel layout. Conditioned on multi-layer DINOv2 features, SLatFlow then generates a PBR Gaussian latent at each active voxel. The structured latent decodes into relightable PBR Gaussians and also serves as input to the mesh decoder, which yields textured meshes with tangent-space normal maps and a complete PBR material set.

Panel of Figure 4 — Figure 4: Effect of multi-layer DINOv2 conditioning. We compare Luce trained with multi-layer DINOv2 features (layers 6, 12, 18, 24) against single-layer DINOv2 features (layer 24 only). Multi-layer conditioning preserves fine spatial detail from the condition image, including legible text and logos on the generated 3D surface. For each variant, the larger shaded render is shown next to a cascade of the per-modality decomposition (albedo, metallic-roughness, surface normals).

Method	Params (B)	Time (s)	Toys4K ( $N\!=\!412$ )								AI-generated images ( $N\!=\!130$ )
Method	Params (B)	Time (s)	FID $\downarrow$	KID $\downarrow$	FID ${}_{\text{dino}}$ $\downarrow$	KID ${}_{\text{dino}}$ $\downarrow$	CLIP $\uparrow$	SigLIP2 $\uparrow$	ULIP $\uparrow$	Uni3D-L $\uparrow$	CLIP $\uparrow$	SigLIP2 $\uparrow$	ULIP $\uparrow$	Uni3D-L $\uparrow$
TRELLIS GS ( Xiang et al., 2025 )	1.70	4.78	30.75	0.212	0.109	0.0013	0.8898	0.9164	—	—	0.8299	0.8339	—	—
TRELLIS mesh ( Xiang et al., 2025 )	1.80	19.67	32.39	0.263	0.145	0.0039	0.8829	0.9092	0.1672	0.3747	0.8000	0.7958	0.1247	0.3278
3DTopia-XL ( Chen et al., 2025 )	1.02	39.61	83.23	2.726	0.542	0.0306	0.7644	0.7810	0.1418	0.2519	0.6481	0.6458	0.1108	0.2197
TRELLIS 2 ( Xiang et al., 2026 )	7.48	199.55	29.22	0.165	0.129	0.0022	0.8895	0.9161	0.1634	0.3635	0.8110	0.8166	0.1207	0.3280
LiTo ( Chang et al., 2026 )	1.84	3.75	29.76	0.208	0.128	0.0025	0.8909	0.9082	—	—	0.8234	0.8240	—	—

Table 1: Image-to-3D generation. We evaluate on two benchmarks: Toys4K (

N\!=\!412

, left) and our 130 AI-generated images (

N\!=\!130

, right). Toys4K reports FID/KID with Inception and DINO ( Oquab et al., 2024 ) backbones (KID and KID

{}_{\text{dino}}

\times 100

); both benchmarks report CLIP ( Radford et al., 2021 ) , SigLIP2 ( Tschannen et al., 2025 ) , ULIP ( Xue et al., 2024 ) , and Uni3D-L ( Zhou et al., 2024 ) . Time(s) is the mean warm wall-clock time per generation on one H100 (Appendix C ). Luce GS renders the decoded PBR Gaussians directly via deferred shading; Luce mesh rows render a textured mesh extracted from the same latent, with and without tangent-space normal map transfer. Best in bold , second-best underlined ; shaded rows are ours. “—” marks metrics that do not apply to that render path; ULIP and Uni3D-L are mesh-only.

Panel of Figure 5 — Figure 5: Image-to-3D generation comparison. Luce preserves legible text and fine surface details on generated 3D assets. For each method, the larger shaded render is shown next to a cascade of the per-modality decomposition (albedo, metallic-roughness, surface normals) when available.

Condition Image PBR GS Render	Condition Image PBR GS Render

Figure 6: Additional generation examples. Luce on four diverse inputs: each cell shows, from left to right, the condition image, the per-modality PBR Gaussians (albedo, metallic-roughness, normals), and a shaded render under the environment map previewed above it.

Method	Color			Albedo			Metallic-Roughness			Normal			Params (B)
Method	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	Params (B)
TRELLIS GS ( Xiang et al., 2025 )	27.0	0.912	0.124	—	—	—	—	—	—	—	—	—	0.17
TRELLIS mesh ( Xiang et al., 2025 )	26.3	0.907	0.118	—	—	—	—	—	—	26.9	0.914	0.113	0.26
3DTopia-XL ( Chen et al., 2025 )	23.9	0.877	0.188	25.4	0.893	0.191	22.6	0.910	0.191	23.3	0.876	0.165	0.03
TRELLIS 2 ( Xiang et al., 2026 )	34.5	0.961	0.074	40.7	0.976	0.051	42.7	0.986	0.038	32.1	0.940	0.093	1.66
LiTo ( Chang et al., 2026 )	31.9	0.944	0.110	—	—	—	—	—	—	27.6	0.912	0.116	0.29

Table 2: Reconstruction quality on a PBR subset of Toys4K (

N\!=\!338

). We report per-modality PSNR, SSIM, and LPIPS for color rendering (combined appearance under fixed illumination), albedo (diffuse color), metallic-roughness, and normal maps (surface orientation). For Luce, the GS row renders the decoded PBR Gaussians directly via deferred shading (no mesh); the mesh rows render a textured mesh from the same latent, with and without baked tangent-space normals. Best in bold , second-best underlined ; shaded rows are ours. “—” indicates that the method does not produce that modality. Params is the autoencoder parameter count in billions.

Panel of Figure 7 — Figure 7: Reconstruction quality on Toys4K. Columns are ground truth and methods; each block shows the modalities that each method reconstructs (albedo, metallic-roughness, normals). Luce preserves fine detail (text, highlights, texture); the green box marks the region magnified below.

Panel of Figure 8 — Figure 8: Tangent-space normal map transfer. Luce bakes tangent-space normals from decoded normal Gaussians onto the extracted mesh, recovering fine surface detail at zero polygon-count cost. Each green box marks the region magnified below.

Appendix figures & tables14 assets

Supplementary material from the paper’s appendix.

Appendix

Architecture
	SLatVAE	SLatFlow
Transformer blocks	16	30
Model channels	1536	1536
I/O block channels	–	[1024]
Attention heads	12	12
MLP ratio	4	6

Appendix

Table 3: SLatVAE and SLatFlow hyperparameters. Architecture diagram in Fig. 9 .

Figure 9: SLatVAE architecture. The encoder compresses the sparse multimodal Gaussian cloud into a compact latent grid. The decoder reconstructs a denser set of Gaussians per voxel per modality, supervised through per-modality differentiable rendering. The same latent also serves as input to a mesh decoder for textured mesh extraction. Full hyperparameters in Table 3 .

Panel of Figure 10 — Figure 10: PBR shaded rendering. The PBR modalities (albedo, metallic-roughness, normals) enable shaded rendering under novel environment maps ( Poly Haven, 2026 ) via standard PBR compositing (Eq. 1 ), without mesh extraction or UV unwrapping. We compare our deferred PBR renderer on the GS fit against the textured mesh rendered with Blender EEVEE ( Blender Online Community, 2024 ) as a reference; the environment maps are listed in Appendix D .

DINOv2 conditioning	Toys4K ( $N\!=\!412$ )						AI-generated images ( $N\!=\!130$ )
DINOv2 conditioning	FID $\downarrow$	KID $\downarrow$	FID ${}_{\text{dino}}$ $\downarrow$	KID ${}_{\text{dino}}$ $\downarrow$	CLIP $\uparrow$	SigLIP2 $\uparrow$	CLIP $\uparrow$	SigLIP2 $\uparrow$
Luce, single-layer (layer 24)	25.21	0.101	0.121	0.0035	0.8977	0.9164	0.8081	0.8143
Luce, multi-layer (layers 6, 12, 18, 24)	20.99	0.033	0.100	0.0028	0.9062	0.9230	0.8519	0.8508

Appendix

Table 4: Multi-layer DINOv2 conditioning ablation. Single-layer (layer 24) versus multi-layer (layers 6, 12, 18, 24) image conditioning for SLatFlow; both rows use the GS render path, so ULIP and Uni3D-L (mesh-only metrics in our setup) are omitted. KID is reported

\times 100

; FID

{}_{\text{dino}}

and KID

{}_{\text{dino}}

use a DINOv2 backbone. Both rows are Luce; the shaded row is our default.

Method	Sparse-structure stage	SLat
3DTopia-XL ( Chen et al., 2025 )	—	25 (DDIM)
TRELLIS ( Xiang et al., 2025 )	25	25
LiTo ( Chang et al., 2026 )	—	20 (Heun)
TRELLIS 2 ( Xiang et al., 2026 )	12	12 (shape) + 12 (texture)
Luce (ours)	12	10

Appendix

Table 5: Inference sampling steps per method. SLat = structured-latent stage. “—” indicates a single-stage method without a separate structure flow.

Condition Image PBR GS Render	Condition Image PBR GS Render

Appendix

Figure 11: Additional generation results. Additional Luce examples across diverse asset categories. Each cell shows, from left to right, the condition image, the per-modality PBR Gaussians (albedo, metallic-roughness, normals), and a shaded render under the environment map previewed above it.

Figure 12: Per-sample baseline comparison (1/8). Rows: methods (top label = condition); Luce rows are ours. Columns: four evaluation views (yaws $300^{\circ}$ , $30^{\circ}$ , $120^{\circ}$ , $210^{\circ}$ ) matching the views used to compute the CLIP and SigLIP2 scores in Table 1 . Per-row metrics show CLIP and SigLIP2, plus ULIP and Uni3D-L for the mesh rows, on the depicted sample. For methods with explicit material decomposition, three small thumbnails are stacked to the right of each shaded render, top to bottom: albedo, metallic-roughness (metallic=R, roughness=G, B unused), and surface normals. LiTo has only the normal thumbnail and TRELLIS GS none; TRELLIS mesh has only a base-color texture, so its uniform metallic-roughness thumbnail is the default material it is rendered with (Appendix D ).

Figure 13: Per-sample baseline comparison (2/8). Same layout as Figure 12 .

Figure 14: Per-sample baseline comparison (3/8). Same layout as Figure 12 .

Figure 15: Per-sample baseline comparison (4/8). Same layout as Figure 12 .

Figure 16: Per-sample baseline comparison (5/8). Same layout as Figure 12 .

Figure 17: Per-sample baseline comparison (6/8). Same layout as Figure 12 .

Figure 18: Per-sample baseline comparison (7/8). Same layout as Figure 12 .

Figure 19: Per-sample baseline comparison (8/8). Same layout as Figure 12 .

Explore similar work

FLUX3D: High-Fidelity 3D Gaussian Generation with Diffusion-Aligned Sparse Representation
Jun 23, 2026Haorui Ji, Weizhe Liu, Hongdong Li +13D Gaussian 3D Generative Models
The Australian National University · ByteDance
GAP3D: Generative Alignment of VLM Latents to Patch-Level Embeddings for 3D Generation
May 27, 2026Polytimi Anna Gkotsi, Andrii Zadaianchuk, Mohammad Mahdi Derakhshani3D Generative Models 3D Assets
University of Amsterdam, The Netherlands
Pixal3D: Pixel-Aligned 3D Generation from Images
May 11, 2026Dong-Yang Li, Wang Zhao, Yuxin Chen +53D Assets 3D Generation
BNRist, Department of Computer Science and Technology, Tsinghua University, China · Tencent ARC Lab, China · Victoria University of Wellington, New Zealand

Jun 23, 2026, cs.CVJ/K move · Enter open · S save

FLUX3D: High-Fidelity 3D Gaussian Generation with Diffusion-Aligned Sparse Representation

Haorui Ji, Weizhe Liu, Hongdong Li, Hengkai Guo

The Australian National University · ByteDance

3D Gaussian 3D Generative Models Voxel 3D Generation Diffusion Alignment Diffusion Transformers+1

Open paper AskPDF arXiv

Sparse voxel representation has emerged as a scalable foundation for image-to-3D Gaussian Splatting (3DGS) generation, yet current methods struggle to preserve high-frequency visual details of input images due to two structural bottlenecks. First, they adopt discriminative 2D features optimized for semantic abstraction to construct sparse voxel latents, which suppress reconstructive cues and induce a representation bottleneck. Second, in the generation stage, standard diffusion transformers lack effective mechanisms to align dense 2D image tokens with sparse 3D voxel latents, resulting in a cross-modal correspondence bottleneck. To address these issues, we propose FLUX3D, a scalable image-to-3DGS framework that boosts both representation learning and cross-modal alignment during generation. We first revisit 2D feature selection for sparse-voxel-based 3D representation learning, propose Diffusion-Aligned Structured Latents (DA-SLAT) and couple it with a decoder-only architecture to improve 3DGS reconstruction fidelity. We also design a sparse-structure-aware diffusion framework, which integrates the Sparse-structure Multimodal Diffusion Transformer (SMDiT) and Modal-Aware Rotary Positional Embedding (MARoPE) to achieve geometry-agnostic 2D-3D alignment. Extensive benchmark experiments demonstrate that FLUX3D yields substantial improvements in appearance fidelity and significantly outperforms all state-of-the-art (SOTA) methods in generating high-quality 3DGS assets.

Abstract

Explore similar work

FLUX3D: High-Fidelity 3D Gaussian Generation with Diffusion-Aligned Sparse Representation

GAP3D: Generative Alignment of VLM Latents to Patch-Level Embeddings for 3D Generation

Pixal3D: Pixel-Aligned 3D Generation from Images