Pith. sign in

REVIEW 5 major objections 6 minor 71 references

FlexPainter: Flexible and Multi-View Consistent Texture Generation

T0 review · 5 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read FlexPainter claims that a single pipeline can take text, images, or both as texture prompts and keep all generated views consistent by synchronizing them in UV space at every diffusion step.

desk verdict Solid systems paper with a genuinely new weighting module, but overclaimed results and an unmeasured depth-alignment premise need fixing before acceptance. read the letter →

arxiv 2506.02620 v1 pith:57VDA4CH submitted 2025-06-03 cs.GR cs.CV

Dongyu Yan , Leyi Wu , Jiantao Lin , Luozhou Wang , Tianshuo Xu , Zhifei Chen , Zhen Yang , Lie Xu
show 2 more authors
Shunsi Zhang Yingcong Chen
This is my paper
classification cs.GRcs.CV
keywords texturegenerationmulti-viewconsistencydiffusionmodelsmulti-modalconditioningclassifier-freeguidanceUV-spacesynchronizationadaptiveweightingstylization
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

FlexPainter sets out to solve two problems in diffusion-based 3D texturing at once: the user is no longer limited to a single prompt modality, and the many camera views used to paint a mesh no longer need to be stitched together from independent generations. The paper proposes one pipeline that accepts text, images, or a weighted combination of both through a shared linear embedding space, and keeps the generated views consistent by synchronizing them in UV space during every diffusion sampling step. It also introduces an image-based classifier-free-guidance trick: feeding a grayscale version of a reference image as the negative prompt strips structural content while preserving color and style, which enables stylization. If the paper is right, a single production pipeline can take a raw mesh and produce seamless, style-controlled, and up to 4K-resolution textures with less manual adjustment.

What carries the argument

The load-bearing object is the shared conditional embedding space built from a text embedder and an image embedder, whose outputs are concatenated into one condition vector that enters the diffusion model through cross-attention. Because that space is linear, the strength and direction of image influence can be manipulated simply, which is what makes modality mixing and stylization possible without retraining. The second central mechanism is the reprojection-based view-synchronization loop: at every denoising step the flow prediction is turned into a clean latent, decoded, split into views, reprojected into UV space, fused by the trainable WeighterNet (which takes partial textures, camera rays, normals, a UV position map, and the noise level $t$), and rasterized back into the multi-view grid, so the next denoising step starts from a single shared UV map. The image-based CFG is the third mechanism: a grayscale reference image, used as a negative prompt, removes structural content while keeping style.

What would settle it

Render a thin or highly detailed mesh (for example, a chair with slats or an animal with fine protrusions) through the depth-conditioned Stage I module, then reproject the generated multi-view images into UV space and compare texel alignment against a ground-truth albedo render; if the reprojection error or normal-consistency score is no better than a heuristic-weight baseline, the depth condition is losing the geometric detail the consistency claim depends on.

Watch

Extended reading notes

Core claim

The central claim is that FlexPainter is a complete texture-generation pipeline whose flexibility and consistency come from two mechanisms. First, text and image prompts are mapped into a shared embedding space where simple linear operations, including concatenation and a scalar weight $\alpha$ per image, let a user interpolate between text and image control, refine an image prompt with text, or apply reference-image stylization. Second, multi-view consistency is attacked at both scales: four surround views are generated together as one $2\times2$ image grid so attention can exchange information across views, and at each denoising step the predicted views are reprojected into UV space, fused by a learned adaptive weighting network (WeighterNet), and rasterized back into the grid so all views are always derived from the same unified texture map. The paper reports that this pipeline outperforms prior text-to-texture and image-to-texture methods on FID, KID, and user preference, and that ablations show the view-synchronization loop, the WeighterNet, and image-based CFG each contribute.

Load-bearing premise

The load-bearing premise is that rendered depth maps retain enough of the mesh's geometric detail to align the Stage I multi-view images with the true surface; the paper itself notes in Supplementary Section D that depth conditions often lose geometric detail, which would make the generated texture fail to fully conform to the mesh.

Editorial extensions

If this is right

  • Users can combine text, one or more images, or both in a single run, with image strength controlled by a scalar $\alpha$, instead of committing to one fixed conditioning modality.
  • Reference-image stylization becomes a negative-prompt operation: grayscaling the reference suppresses its structure and layout while transferring color and tone.
  • Generating four views as one grid and syncing them through UV space at every sampling step should reduce both ghosting artifacts and the Janus problem that plague single-view or unmerged multi-view approaches.
  • The trained WeighterNet replaces fixed cosine-weight blending, so partial-texture fusion adapts to different geometries, views, and noise levels.
  • With the completion and enhancement stages, the pipeline can output full 4K texture maps that render without visible seams.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Because the embedding space is linear, $\alpha$-interpolation between two conflicting reference images could serve as a controllable blend dial; the paper does not test this, and a quantitative style-content metric would be a natural extension.
  • Given the acknowledged depth-detail loss, feeding Stage I higher-fidelity geometric cues such as normal maps or sharper depth should reduce UV reprojection error; this is a testable extension, not a paper claim.
  • The WeighterNet's inputs (rays, normals, position map, timestep) are not texture-specific, so it could plausibly be reused as a general multi-view fusion module for other image-to-3D tasks, though the authors do not claim this.
  • The grayscale-negative-prompt stylization may fail when the reference image's style is itself structural, such as hatching or line art, because suppressing structure would erase the desired style; testing that boundary would define the method's operating range.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 6 minor

Summary. FlexPainter proposes a texture generation pipeline for 3D meshes that supports text, image, and combined text+image conditioning through a shared embedding space, an image-based CFG mechanism for stylization, multi-view grid generation with depth conditioning, reprojection-based view synchronization with a learned WeighterNet, and TEXGen/Real-ESRGAN based completion and enhancement. The method is evaluated for text-to-texture on 100 unseen GSO objects against TEXTure, Text2Tex, SyncMVD, Paint3D, and TEXGen using FID, KID, and a user study, and for image-to-texture against Paint3D. Ablations isolate view synchronization, WeighterNet, and image-based CFG. The paper claims that FlexPainter significantly outperforms state-of-the-art methods in both flexibility and generation quality.

Significance. If the claims are substantiated, FlexPainter would be a practical contribution to texture generation, combining global multi-view consistency via a 2x2 image grid, local consistency via UV-space synchronization with an adaptive weighting network, and flexible multi-modal conditioning in a single system. The paper's strengths include evaluation on unseen GSO objects, ablations of the main modules, a user study, and a broad qualitative application set. However, the evaluation does not yet establish the strongest claims: the geometric alignment between generated textures and the target mesh is not measured, the KID numbers in Table 1 partially contradict the headline quality claim, and several core mechanisms are asserted rather than verified. These gaps are load-bearing for the paper's central claims.

major comments (5)
  1. [Section 5.2, Table 1] The abstract's claim that FlexPainter 'significantly outperforms state-of-the-art methods in both flexibility and generation quality' is not fully supported by Table 1. FlexPainter achieves the best FID (71.621) and user preference (28.3%), but its KID (58.465) is worse than SyncMVD's KID (41.737). Moreover, no confidence intervals or significance tests are reported, so the FID gap to TEXGen (71.621 vs. 72.900) and the user-preference gaps may fall within sampling noise. Please report uncertainty estimates for all metrics and temper the claim accordingly.
  2. [Section 4.2 and Supplementary Section D] The central 'highly consistent texture generation' claim requires that the generated multi-view images align with the target mesh, but this premise is not quantitatively validated. The view-synchronization mechanism (Eqs. 5-9) reprojects decoded latent images into UV space using mesh depth and re-rasterizes them; if the depth condition does not constrain generated appearance to the surface boundaries, texels are pulled from incorrect image locations and averaging across views will blur or float features. Supplementary D explicitly concedes that depth-map conditioning 'often results in the loss of geometric detail from the original mesh' and that Stage I results 'may not fully align with the original mesh.' FID, KID, and user preference measure overall visual quality and prompt adherence, not geometric alignment. Please add a direct alignment metric (e.g., silhouette IoU between generated views and rendered depths, or depth/boundary consistency of the final texture) and report failure cases.
  3. [Section 4.1, Eq. (4)] The flexible-conditioning contribution rests on the assertion that T5 text embeddings and Redux image embeddings live in a shared linear space in which weighted addition preserves semantics. This is stated rather than derived or verified: no experiment isolates Eq. (4), and the claim that 'even with single-modal training, the linear structure and combination property can be well preserved' has no supporting evidence. Given that the interpolation and text-guided refinement results in Figs. 2 and 6 are only qualitative, please add a quantitative or at least a controlled ablation of the linear mixing rule (e.g., compare Eq. (4) with concatenation, or measure prompt-following on a set of interpolation examples).
  4. [Section 5.3, Table 2] The image-to-texture comparison is limited to Paint3D, even though the paper's contribution is multi-modal conditioning. To establish 'significantly outperforms state-of-the-art methods in flexibility,' the authors should compare against other image-conditioned texture generation methods, such as FlexiTex [15] and StyleTex [17], using the same evaluation protocol. Without such baselines, the advantage of FlexPainter's image-based conditional guidance is not demonstrated.
  5. [Section 4.1, Image-based CFG] The claim that a grayscaled reference image used as a CFG negative prompt 'eliminates structural information while preserving stylistic features' is a key mechanism for stylization but is only supported by qualitative examples. Please provide a focused experiment showing that the grayscale negative, rather than the choice of reference image or other CFG settings, is responsible for the content-style disentanglement, and report quantitative stylization metrics if available.
minor comments (6)
  1. [Section 4.3] The sentence 'At can be produced.' is a fragment and should be completed or removed.
  2. [Figure 9] The caption reads 'More cases of our imgae-to-texture generation'; 'imgae' should be 'image'.
  3. [Section 5.1] The text 'as our comparison baseline In particular, we use' is missing a period after 'baseline'; please fix the punctuation.
  4. [Supplementary A.2] The trigger-prompt line contains unbalanced quotation marks ('We use "'a grid of 2x2 multi-view image. white background."'); please clean up the quoting.
  5. [Supplementary D] The limitations section refers to 'Stage I' and 'Stage II' but these stages are not explicitly defined in the main text; please define them at first use.
  6. [General] No code or model release is mentioned; given the pipeline's multiple trained components (LoRA, WeighterNet, completion, enhancement), a code/data release statement would greatly improve reproducibility.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the pipeline is trained and evaluated on external data; the geometric-alignment caveat is a validation gap, not a circular derivation.

full rationale

No circular step can be exhibited. FlexPainter's conditioning, view-synchronization, weighting, completion, and enhancement modules are built from external pretrained models (FLUX.1-dev, T5, Redux, TEXGen, Real-ESRGAN) or are trained with ground-truth supervision and then evaluated on 100 unseen GSO objects with FID, KID, and user preference. The reprojection-based view synchronization in Eqs. (5)-(9) is a mechanism that re-derives all views from one fused UV map; the paper does not present this as a prediction, and its benefit is ablated in Table 3. The WeighterNet is trained on ground-truth textures paired with simulated denoised textures, which is standard supervised learning rather than a fitted parameter disguised as a result. The only overlapping-author citation is FlexGen [36], which appears in Related Work as one of several prior multi-view generation methods; no FlexPainter result depends on FlexGen outputs, so the self-citation is not load-bearing. The paper's own Supplementary Section D states that 'utilizing depth maps as geometric conditions... often results in the loss of geometric detail from the original mesh' and that Stage I results may 'not fully aligning with the original mesh'; this is an honest limitation and an unvalidated geometric-alignment premise, and no quantitative alignment metric is reported, but this is a correctness/validation concern rather than circularity because no equation or derived result is defined in terms of the consistency claim it is supposed to support. Similarly, the claims about linear embedding-space manipulation and grayscale-negative CFG are asserted without formal proof, but they are empirically testable and not tautological. The derivation chain is therefore self-contained against external benchmarks, and the minor self-citation plus the acknowledged alignment gap justify only a low score.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The method relies on strong priors from FLUX.1-dev, T5, Redux, TEXGen, and Real-ESRGAN without re-deriving them. The paper-specific assumptions (embedding-space linearity, grayscale-negative style transfer, simulation-to-real transfer for WeighterNet) are asserted and only indirectly evaluated. The most fragile assumption, depth-based geometry alignment, is explicitly acknowledged as a limitation in the supplementary material.

free parameters (4)
  • Interpolation weights alpha_i in Eq. (4) = User-specified, 0 or 1 in experiments, continuous in Fig. 2
    Controls mix of text and image conditions; central to the flexible-aggregation claim, but linearity is asserted, not validated quantitatively.
  • CFG scales = distilled=6, explicit=2
    Hand-chosen in Appendix A.5 for all experiments; no sensitivity study.
  • Negative embedding for explicit CFG = white image embedding
    Hand-chosen in Appendix A.5; effect on quality not analyzed.
  • LoRA trigger prompt = 'a grid of 2x2 multi-view image. white background.'
    Hand-written prompt for multi-view grid generation; prompt sensitivity untested.
assumptions (6)
  • standard math Rectified-flow velocity field v_t in Eq. (1) and the clean-data estimate x_t^0 = x_t - t v_t(x_t) are valid at every denoising step.
    This follows from the definition of rectified flow; used in Eq. (5) for view synchronization.
  • domain assumption The VAE encode/decode cycle (E and D) is approximately lossless, so repeated decode-reproject-encode steps in Eqs. (6)-(8) do not accumulate distortions that break the diffusion manifold.
    The view synchronization loop performs multiple encode/decode operations per step; the paper does not analyze distortion accumulation and even notes in Sec. 5.4 that encode-decode cycles degrade text-CFG quality.
  • ad hoc to paper T5 text embeddings and Redux image embeddings share a linear space in which weighted addition preserves semantic meaning (Eq. 4).
    The paper asserts this linearity ('our embedding space retains the superior properties of a linear space') without proof or quantitative validation beyond qualitative interpolation examples.
  • ad hoc to paper A grayscaled reference image used as a CFG negative prompt removes structural information while preserving style.
    Heuristic central to the stylization claim; supported only by qualitative examples and one ablation for image-CFG vs text-CFG, not a dedicated style-transfer evaluation.
  • domain assumption WeighterNet trained on simulated partial textures (ground-truth textures plus noise) generalizes to actual partial textures produced by the multi-view model at inference.
    Training pairs simulated per Sec. 5.1; the distribution shift from simulation to real generation is not quantified.
  • domain assumption Rendered depth maps capture sufficient geometric detail to align generated views with the mesh.
    Limitations section states this often fails: 'loss of geometric detail... can lead to the generation results in Stage I not fully aligning with the original mesh'.

how reviews work

0 comments
Cite this review

Pith. "Pith review of FlexPainter: Flexible and Multi-View Consistent Texture Generation." pith.science (2026). https://pith.science/paper/57VDA4CH

@misc{pith2026250602620,
  author       = {Pith},
  title        = {Pith review of: FlexPainter: Flexible and Multi-View Consistent Texture Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/57VDA4CH}},
  note         = {Machine review of arXiv:2506.02620}
}
read the original abstract

Texture map production is an important part of 3D modeling and determines the rendering quality. Recently, diffusion-based methods have opened a new way for texture generation. However, restricted control flexibility and limited prompt modalities may prevent creators from producing desired results. Furthermore, inconsistencies between generated multi-view images often lead to poor texture generation quality. To address these issues, we introduce \textbf{FlexPainter}, a novel texture generation pipeline that enables flexible multi-modal conditional guidance and achieves highly consistent texture generation. A shared conditional embedding space is constructed to perform flexible aggregation between different input modalities. Utilizing such embedding space, we present an image-based CFG method to decompose structural and style information, achieving reference image-based stylization. Leveraging the 3D knowledge within the image diffusion prior, we first generate multi-view images simultaneously using a grid representation to enhance global understanding. Meanwhile, we propose a view synchronization and adaptive weighting module during diffusion sampling to further ensure local consistency. Finally, a 3D-aware texture completion model combined with a texture enhancement model is used to generate seamless, high-resolution texture maps. Comprehensive experiments demonstrate that our framework significantly outperforms state-of-the-art methods in both flexibility and generation quality.

Figures

Figures reproduced from arXiv: 2506.02620 by the authors.

Figure 1
Figure 1. FlexPainter generates diverse, high-quality textures based on various flexible user prompts. Abstract Texture map production is an important part of 3D modeling and determines the rendering quality. Recently, diffusion￾based methods have opened a new way for texture gen￾eration. However, restricted control flexibility and limited prompt modalities may prevent creators from producing de￾sired results. Furthermore, in… view at source ↗
Figure 2
Figure 2. Example of modality aggregation and semantic manipu [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Pipeline of our method. We first generate multi-view images using the conditional input from the user. The top two cases show [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Qualitative results on text-to-texture generation. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Qualitative results on image-to-texture generation. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Our Applications includes tasks of text-to-texture, image-to-texture, text-guided image refinement, and reference image-based [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Qualitative results of our ablation study. Our full model [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: More cases of our text-to-texture generation. [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]
Figure 9
Figure 9. Figure 9: More cases of our imgae-to-texture generation. [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]
Figure 10
Figure 10. Figure 10: More cases of our applications. 13 [PITH_FULL_IMAGE:figures/full_fig_p013_10.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

71 extracted references · 29 canonical work pages

  1. [15]

    Flexitex: Enhancing tex- ture generation with visual guidance.arXiv preprint arXiv:2409.12431, 2024

    DaDong Jiang, Xianghui Yang, Zibo Zhao, Sheng Zhang, Jiaao Yu, Zeqiang Lai, Shaoxiong Yang, Chunchao Guo, Xiaobo Zhou, and Zhihui Ke. Flexitex: Enhancing tex- ture generation with visual guidance.arXiv preprint arXiv:2409.12431, 2024. 2, 3

    arXiv 2024

  2. [17]

    Styletex: Style image-guided texture generation for 3d models.ACM Trans- actions on Graphics (TOG), 43(6):1–14, 2024

    Zhiyu Xie, Yuqing Zhang, Xiangjun Tang, Yiqian Wu, De- han Chen, Gongsheng Li, and Xiaogang Jin. Styletex: Style image-guided texture generation for 3d models.ACM Trans- actions on Graphics (TOG), 43(6):1–14, 2024. 2, 3

    work page 2024

  3. [1]

    Photorealistic text-to-image diffusion models with deep language understanding.Advances in neural information processing systems, 35:36479–36494, 2022

    Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily L Denton, Kamyar Ghasemipour, Raphael Gontijo Lopes, Burcu Karagol Ayan, Tim Salimans, et al. Photorealistic text-to-image diffusion models with deep language understanding.Advances in neural information processing systems, 35:36479–36494, 2022. 2

    2022

  4. [2]

    Hierarchical text-conditional image gen- eration with clip latents.arXiv preprint arXiv:2204.06125, 1(2):3, 2022

    Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen. Hierarchical text-conditional image gen- eration with clip latents.arXiv preprint arXiv:2204.06125, 1(2):3, 2022

    arXiv 2022

  5. [3]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer. High-resolution image synthesis with latent diffusion models. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022

    2022

  6. [4]

    Sdxl: Improving latent diffusion mod- els for high-resolution image synthesis.arXiv preprint arXiv:2307.01952, 2023

    Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas M ¨uller, Joe Penna, and Robin Rombach. Sdxl: Improving latent diffusion mod- els for high-resolution image synthesis.arXiv preprint arXiv:2307.01952, 2023

    arXiv 2023

  7. [5]

    Scaling recti- fied flow transformers for high-resolution image synthesis

    Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas M ¨uller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, et al. Scaling recti- fied flow transformers for high-resolution image synthesis. InForty-first international conference on machine learning, 2024

    2024

  8. [6]

    Flux.https:// github.com/ black-forest-labs/flux, 2023

    Black-Forest-Labs. Flux.https:// github.com/ black-forest-labs/flux, 2023. 2, 6, 14

    work page 2023

Show all 71 references
  1. [7]

    Texture: Text-guided texturing of 3d shapes

    Elad Richardson, Gal Metzer, Yuval Alaluf, Raja Giryes, and Daniel Cohen-Or. Texture: Text-guided texturing of 3d shapes. InACM SIGGRAPH 2023 conference proceedings, pages 1–11, 2023. 2, 3, 6

    2023

  2. [8]

    Text2tex: Text-driven tex- ture synthesis via diffusion models

    Dave Zhenyu Chen, Yawar Siddiqui, Hsin-Ying Lee, Sergey Tulyakov, and Matthias Nießner. Text2tex: Text-driven tex- ture synthesis via diffusion models. InProceedings of the 8 IEEE/CVF International Conference on Computer Vision, pages 18558–18568, 2023. 2, 6

    2023

  3. [9]

    Texro: Generating delicate textures of 3d models by recursive optimization.arXiv preprint arXiv:2403.15009, 2024

    Jinbo Wu, Xing Liu, Chenming Wu, Xiaobo Gao, Jialun Liu, Xinqi Liu, Chen Zhao, Haocheng Feng, Errui Ding, and Jingdong Wang. Texro: Generating delicate textures of 3d models by recursive optimization.arXiv preprint arXiv:2403.15009, 2024. 3

    2024 arXiv

  4. [10]

    Make-a-texture: Fast shape-aware texture genera- tion in 3 seconds.arXiv preprint arXiv:2412.07766, 2024

    Xiaoyu Xiang, Liat Sless Gorelik, Yuchen Fan, Omri Arm- strong, Forrest Iandola, Yilei Li, Ita Lifshitz, and Rakesh Ranjan. Make-a-texture: Fast shape-aware texture genera- tion in 3 seconds.arXiv preprint arXiv:2412.07766, 2024. 3

    2024 arXiv

  5. [11]

    Genesistex: Adapting image denoising diffu- sion to texture space

    Chenjian Gao, Boyan Jiang, Xinghui Li, Yingpeng Zhang, and Qian Yu. Genesistex: Adapting image denoising diffu- sion to texture space. InProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 4620–4629, 2024. 3

    2024

  6. [12]

    Tex- gen: Text-guided 3d texture generation with multi-view sam- pling and resampling

    Dong Huo, Zixin Guo, Xinxin Zuo, Zhihao Shi, Juwei Lu, Peng Dai, Songcen Xu, Li Cheng, and Yee-Hong Yang. Tex- gen: Text-guided 3d texture generation with multi-view sam- pling and resampling. InEuropean Conference on Computer Vision, pages 352–368. Springer, 2025. 2, 3

    2025

  7. [13]

    Mvpaint: Synchronized multi-view diffusion for paint- ing anything 3d.arXiv preprint arXiv:2411.02336, 2024

    Wei Cheng, Juncheng Mu, Xianfang Zeng, Xin Chen, Anqi Pang, Chi Zhang, Zhibin Wang, Bin Fu, Gang Yu, Ziwei Liu, et al. Mvpaint: Synchronized multi-view diffusion for paint- ing anything 3d.arXiv preprint arXiv:2411.02336, 2024. 2, 3

    2024 arXiv

  8. [14]

    Paint3d: Paint anything 3d with lighting-less texture diffusion models

    Xianfang Zeng, Xin Chen, Zhongqi Qi, Wen Liu, Zibo Zhao, Zhibin Wang, Bin Fu, Yong Liu, and Gang Yu. Paint3d: Paint anything 3d with lighting-less texture diffusion models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4252–4262, 20...

    2024

  9. [16]

    Ip- adapter: Text compatible image prompt adapter for text-to- image diffusion models.arXiv preprint arXiv:2308.06721,

    Hu Ye, Jun Zhang, Sibo Liu, Xiao Han, and Wei Yang. Ip- adapter: Text compatible image prompt adapter for text-to- image diffusion models.arXiv preprint arXiv:2308.06721,

    arXiv

  10. [18]

    Text-guided texturing by synchronized multi-view diffusion

    Yuxin Liu, Minshan Xie, Hanyuan Liu, and Tien-Tsin Wong. Text-guided texturing by synchronized multi-view diffusion. InSIGGRAPH Asia 2024 Conference Papers, pages 1–11,

    2024

  11. [19]

    Texfusion: Synthesizing 3d textures with text-guided image diffusion models

    Tianshi Cao, Karsten Kreis, Sanja Fidler, Nicholas Sharp, and Kangxue Yin. Texfusion: Synthesizing 3d textures with text-guided image diffusion models. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 4169–4181, 2023. 2, 3

    2023

  12. [20]

    Meta 3d texturegen: Fast and consistent texture generation for 3d objects.arXiv preprint arXiv:2407.02430, 2024

    Raphael Bensadoun, Yanir Kleiman, Idan Azuri, Omri Harosh, Andrea Vedaldi, Natalia Neverova, and Oran Gafni. Meta 3d texturegen: Fast and consistent texture generation for 3d objects.arXiv preprint arXiv:2407.02430, 2024. 2, 3

    2024 arXiv

  13. [21]

    Classifier-free diffusion guidance.arXiv preprint arXiv:2207.12598, 2022

    Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance.arXiv preprint arXiv:2207.12598, 2022. 2, 4, 14

    2022 arXiv

  14. [22]

    Texpainter: Generative mesh texturing with multi-view consistency

    Hongkun Zhang, Zherong Pan, Congyi Zhang, Lifeng Zhu, and Xifeng Gao. Texpainter: Generative mesh texturing with multi-view consistency. InACM SIGGRAPH 2024 Confer- ence Papers, pages 1–11, 2024. 2, 3

    2024

  15. [23]

    Zero-1-to- 3: Zero-shot one image to 3d object

    Ruoshi Liu, Rundi Wu, Basile Van Hoorick, Pavel Tok- makov, Sergey Zakharov, and Carl V ondrick. Zero-1-to- 3: Zero-shot one image to 3d object. InProceedings of the IEEE/CVF international conference on computer vision, pages 9298–9309, 2023. 2

    2023

  16. [24]

    Consistent-1-to-3: Consistent image to 3d view syn- thesis via geometry-aware diffusion models

    Jianglong Ye, Peng Wang, Kejie Li, Yichun Shi, and Heng Wang. Consistent-1-to-3: Consistent image to 3d view syn- thesis via geometry-aware diffusion models. In2024 Inter- national Conference on 3D Vision (3DV), pages 664–674. IEEE, 2024. 2

    2024

  17. [25]

    Zero123++: a single image to consistent multi-view dif- fusion base model.arXiv preprint arXiv:2310.15110, 2023

    Ruoxi Shi, Hansheng Chen, Zhuoyang Zhang, Minghua Liu, Chao Xu, Xinyue Wei, Linghao Chen, Chong Zeng, and Hao Su. Zero123++: a single image to consistent multi-view dif- fusion base model.arXiv preprint arXiv:2310.15110, 2023. 2

    arXiv

  18. [26]

    Mvdream: Multi-view diffusion for 3d gen- eration.arXiv preprint arXiv:2308.16512, 2023

    Yichun Shi, Peng Wang, Jianglong Ye, Mai Long, Kejie Li, and Xiao Yang. Mvdream: Multi-view diffusion for 3d gen- eration.arXiv preprint arXiv:2308.16512, 2023. 2

    2023 arXiv

  19. [27]

    Instant-3d: Instant neural radiance field training towards on- device ar/vr 3d reconstruction

    Sixu Li, Chaojian Li, Wenbo Zhu, Boyang Yu, Yang Zhao, Cheng Wan, Haoran You, Huihong Shi, and Yingyan Lin. Instant-3d: Instant neural radiance field training towards on- device ar/vr 3d reconstruction. InProceedings of the 50th Annual International Symposium on Computer Archi...

    2023

  20. [28]

    Imagedream: Image-prompt multi-view diffusion for 3d generation.arXiv preprint arXiv:2312.02201, 2023

    Peng Wang and Yichun Shi. Imagedream: Image-prompt multi-view diffusion for 3d generation.arXiv preprint arXiv:2312.02201, 2023

    2023 arXiv

  21. [29]

    Wonder3d: Sin- gle image to 3d using cross-domain diffusion

    Xiaoxiao Long, Yuan-Chen Guo, Cheng Lin, Yuan Liu, Zhiyang Dou, Lingjie Liu, Yuexin Ma, Song-Hai Zhang, Marc Habermann, Christian Theobalt, et al. Wonder3d: Sin- gle image to 3d using cross-domain diffusion. InProceed- ings of the IEEE/CVF conference on computer vision and pat...

    2024

  22. [30]

    Unique3d: High-quality and efficient 3d mesh generation from a single image

    Kailu Wu, Fangfu Liu, Zhihan Cai, Runjie Yan, Hanyang Wang, Yating Hu, Yueqi Duan, and Kaisheng Ma. Unique3d: High-quality and efficient 3d mesh generation from a single image. InThe Thirty-eighth Annual Conference on Neural Information Processing Systems, 2024. 2

    2024

  23. [31]

    Sv3d: Novel multi-view syn- thesis and 3d generation from a single image using latent video diffusion

    Vikram V oleti, Chun-Han Yao, Mark Boss, Adam Letts, David Pankratz, Dmitry Tochilkin, Christian Laforte, Robin Rombach, and Varun Jampani. Sv3d: Novel multi-view syn- thesis and 3d generation from a single image using latent video diffusion. InEuropean Conference on Computer ...

    2025

  24. [32]

    Dmv3d: Denoising multi-view diffu- sion using 3d large reconstruction model.arXiv preprint arXiv:2311.09217, 2023

    Yinghao Xu, Hao Tan, Fujun Luan, Sai Bi, Peng Wang, Ji- ahao Li, Zifan Shi, Kalyan Sunkavalli, Gordon Wetzstein, Zexiang Xu, et al. Dmv3d: Denoising multi-view diffu- sion using 3d large reconstruction model.arXiv preprint arXiv:2311.09217, 2023. 9

    2023 arXiv

  25. [33]

    Im-3d: Iterative multiview diffusion and re- construction for high-quality 3d generation.arXiv preprint arXiv:2402.08682, 2024

    Luke Melas-Kyriazi, Iro Laina, Christian Rupprecht, Na- talia Neverova, Andrea Vedaldi, Oran Gafni, and Filippos Kokkinos. Im-3d: Iterative multiview diffusion and re- construction for high-quality 3d generation.arXiv preprint arXiv:2402.08682, 2024

    2024 arXiv

  26. [34]

    Syncdreamer: Gen- erating multiview-consistent images from a single-view im- age.arXiv preprint arXiv:2309.03453, 2023

    Yuan Liu, Cheng Lin, Zijiao Zeng, Xiaoxiao Long, Lingjie Liu, Taku Komura, and Wenping Wang. Syncdreamer: Gen- erating multiview-consistent images from a single-view im- age.arXiv preprint arXiv:2309.03453, 2023

    2023 arXiv

  27. [35]

    Consistent123: Improve consistency for one image to 3d object synthesis.arXiv preprint arXiv:2310.08092, 2023

    Haohan Weng, Tianyu Yang, Jianan Wang, Yu Li, Tong Zhang, CL Chen, and Lei Zhang. Consistent123: Improve consistency for one image to 3d object synthesis.arXiv preprint arXiv:2310.08092, 2023

    2023 arXiv

  28. [36]

    Flex- gen: Flexible multi-view generation from text and image in- puts.arXiv preprint arXiv:2410.10745, 2024

    Xinli Xu, Wenhang Ge, Jiantao Lin, Jiawei Feng, Lie Xu, HanFeng Zhao, Shunsi Zhang, and Ying-Cong Chen. Flex- gen: Flexible multi-view generation from text and image in- puts.arXiv preprint arXiv:2410.10745, 2024. 2

    2024 arXiv

  29. [37]

    Solid tex- ture synthesis from 2d exemplars

    Johannes Kopf, Chi-Wing Fu, Daniel Cohen-Or, Oliver Deussen, Dani Lischinski, and Tien-Tsin Wong. Solid tex- ture synthesis from 2d exemplars. InACM SIGGRAPH 2007 papers, pages 2–es. 2007. 3

    2007

  30. [38]

    Appearance-space texture synthesis.ACM Transactions on Graphics (TOG), 25(3):541–548, 2006

    Sylvain Lefebvre and Hugues Hoppe. Appearance-space texture synthesis.ACM Transactions on Graphics (TOG), 25(3):541–548, 2006

    2006

  31. [39]

    Texture synthesis on surfaces

    Greg Turk. Texture synthesis on surfaces. InProceedings of the 28th annual conference on Computer graphics and inter- active techniques, pages 347–354, 2001

    2001

  32. [40]

    State of the art in example-based texture synthesis.Euro- graphics 2009, State of the Art Report, EG-STAR, pages 93– 117, 2009

    Li-Yi Wei, Sylvain Lefebvre, Vivek Kwatra, and Greg Turk. State of the art in example-based texture synthesis.Euro- graphics 2009, State of the Art Report, EG-STAR, pages 93– 117, 2009. 3

    2009

  33. [41]

    Texture fields: Learning tex- ture representations in function space

    Michael Oechsle, Lars Mescheder, Michael Niemeyer, Thilo Strauss, and Andreas Geiger. Texture fields: Learning tex- ture representations in function space. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 4531–4540, 2019. 3

    2019

  34. [42]

    Texturify: Generating textures on 3d shape surfaces

    Yawar Siddiqui, Justus Thies, Fangchang Ma, Qi Shan, Matthias Nießner, and Angela Dai. Texturify: Generating textures on 3d shape surfaces. InEuropean Conference on Computer Vision, pages 72–88. Springer, 2022

    2022

  35. [43]

    Mesh2tex: Generating mesh textures from image queries

    Alexey Bokhovkin, Shubham Tulsiani, and Angela Dai. Mesh2tex: Generating mesh textures from image queries. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 8918–8928, 2023

    2023

  36. [44]

    Tuvf: Learning generalizable texture uv radiance fields

    An-Chieh Cheng, Xueting Li, Sifei Liu, and Xiaolong Wang. Tuvf: Learning generalizable texture uv radiance fields. arXiv preprint arXiv:2305.03040, 2023. 3

    2023 arXiv

  37. [45]

    Generative adversarial networks.Commu- nications of the ACM, 63(11):139–144, 2020

    Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial networks.Commu- nications of the ACM, 63(11):139–144, 2020. 3

    2020

  38. [46]

    Analyzing and improv- ing the image quality of stylegan

    Tero Karras, Samuli Laine, Miika Aittala, Janne Hellsten, Jaakko Lehtinen, and Timo Aila. Analyzing and improv- ing the image quality of stylegan. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8110–8119, 2020. 3

    2020

  39. [47]

    Dreamfusion: Text-to-3d using 2d diffusion.arXiv preprint arXiv:2209.14988, 2022

    Ben Poole, Ajay Jain, Jonathan T Barron, and Ben Milden- hall. Dreamfusion: Text-to-3d using 2d diffusion.arXiv preprint arXiv:2209.14988, 2022. 3

    2022 arXiv

  40. [48]

    Magic3d: High-resolution text-to-3d content creation

    Chen-Hsuan Lin, Jun Gao, Luming Tang, Towaki Takikawa, Xiaohui Zeng, Xun Huang, Karsten Kreis, Sanja Fidler, Ming-Yu Liu, and Tsung-Yi Lin. Magic3d: High-resolution text-to-3d content creation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition...

    2023

  41. [49]

    Fan- tasia3d: Disentangling geometry and appearance for high- quality text-to-3d content creation

    Rui Chen, Yongwei Chen, Ningxin Jiao, and Kui Jia. Fan- tasia3d: Disentangling geometry and appearance for high- quality text-to-3d content creation. InProceedings of the IEEE/CVF international conference on computer vision, pages 22246–22256, 2023

    2023

  42. [50]

    Latent-nerf for shape-guided generation of 3d shapes and textures

    Gal Metzer, Elad Richardson, Or Patashnik, Raja Giryes, and Daniel Cohen-Or. Latent-nerf for shape-guided generation of 3d shapes and textures. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12663–12673, 2023

    2023

  43. [51]

    Texture- dreamer: Image-guided texture synthesis through geometry- aware diffusion

    Yu-Ying Yeh, Jia-Bin Huang, Changil Kim, Lei Xiao, Thu Nguyen-Phuoc, Numair Khan, Cheng Zhang, Manmohan Chandraker, Carl S Marshall, Zhao Dong, et al. Texture- dreamer: Image-guided texture synthesis through geometry- aware diffusion. InProceedings of the IEEE/CVF Confer- ence...

    2024

  44. [52]

    An image is worth one word: Personalizing text-to- image generation using textual inversion.arXiv preprint arXiv:2208.01618, 2022

    Rinon Gal, Yuval Alaluf, Yuval Atzmon, Or Patash- nik, Amit H Bermano, Gal Chechik, and Daniel Cohen- Or. An image is worth one word: Personalizing text-to- image generation using textual inversion.arXiv preprint arXiv:2208.01618, 2022. 3

    2022 arXiv

  45. [53]

    Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation

    Nataniel Ruiz, Yuanzhen Li, Varun Jampani, Yael Pritch, Michael Rubinstein, and Kfir Aberman. Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 22500–...

    2023

  46. [54]

    Learning transferable visual models from natural language supervi- sion

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervi- sion. InInternational conference on machine learning, p...

    2021

  47. [55]

    Vcd-texture: Variance alignment based 3d-2d co- denoising for text-guided texturing

    Shang Liu, Chaohui Yu, Chenjie Cao, Wen Qian, and Fan Wang. Vcd-texture: Variance alignment based 3d-2d co- denoising for text-guided texturing. InEuropean Conference on Computer Vision, pages 373–389. Springer, 2024. 3

    2024

  48. [56]

    Flashtex: Fast relightable mesh texturing with lightcontrolnet

    Kangle Deng, Timothy Omernick, Alexander Weiss, Deva Ramanan, Jun-Yan Zhu, Tinghui Zhou, and Maneesh Agrawala. Flashtex: Fast relightable mesh texturing with lightcontrolnet. InEuropean Conference on Computer Vi- sion, pages 90–107. Springer, 2025. 3

    2025

  49. [57]

    Texture generation on 3d meshes with point- uv diffusion

    Xin Yu, Peng Dai, Wenbo Li, Lan Ma, Zhengzhe Liu, and Xiaojuan Qi. Texture generation on 3d meshes with point- uv diffusion. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 4206–4216, 2023. 3

    2023

  50. [58]

    Texgen: a generative diffusion model for mesh textures

    Xin Yu, Ze Yuan, Yuan-Chen Guo, Ying-Tian Liu, Jianhui Liu, Yangguang Li, Yan-Pei Cao, Ding Liang, and Xiaojuan 10 Qi. Texgen: a generative diffusion model for mesh textures. ACM Transactions on Graphics (TOG), 43(6):1–14, 2024. 3, 6, 14

    2024

  51. [59]

    Flow matching for generative mod- eling.arXiv preprint arXiv:2210.02747, 2022

    Yaron Lipman, Ricky TQ Chen, Heli Ben-Hamu, Maximil- ian Nickel, and Matt Le. Flow matching for generative mod- eling.arXiv preprint arXiv:2210.02747, 2022. 3

    2022 arXiv

  52. [60]

    Auto-encoding vari- ational bayes, 2013

    Diederik P Kingma, Max Welling, et al. Auto-encoding vari- ational bayes, 2013. 3

    2013

  53. [61]

    Lora: Low-rank adaptation of large language models.arXiv preprint arXiv:2106.09685, 2021

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen- Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. Lora: Low-rank adaptation of large language models.arXiv preprint arXiv:2106.09685, 2021. 4, 6

    2021 arXiv

  54. [62]

    Real-esrgan: Training real-world blind super-resolution with pure synthetic data

    Xintao Wang, Liangbin Xie, Chao Dong, and Ying Shan. Real-esrgan: Training real-world blind super-resolution with pure synthetic data. InProceedings of the IEEE/CVF inter- national conference on computer vision, pages 1905–1914,

    1905

  55. [63]

    Exploring the limits of transfer learning with a unified text-to-text transformer.Journal of machine learning research, 21(140):1–67, 2020

    Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. Exploring the limits of transfer learning with a unified text-to-text transformer.Journal of machine learning research, 21(140):1–67, 2020. 6

    2020

  56. [64]

    An image is worth 16x16 words: Trans- formers for image recognition at scale.arXiv preprint arXiv:2010.11929, 2020

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, et al. An image is worth 16x16 words: Trans- formers for image recognition at scale.arXiv preprint ar...

    2010 arXiv

  57. [65]

    Point transformer v3: Simpler faster stronger

    Xiaoyang Wu, Li Jiang, Peng-Shuai Wang, Zhijian Liu, Xi- hui Liu, Yu Qiao, Wanli Ouyang, Tong He, and Hengshuang Zhao. Point transformer v3: Simpler faster stronger. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4840–4851, 2024. 6

    2024

  58. [66]

    Objaverse: A universe of annotated 3d objects

    Matt Deitke, Dustin Schwenk, Jordi Salvador, Luca Weihs, Oscar Michel, Eli VanderBilt, Ludwig Schmidt, Kiana Ehsani, Aniruddha Kembhavi, and Ali Farhadi. Objaverse: A universe of annotated 3d objects. InProceedings of the IEEE/CVF conference on computer vision and pattern reco...

    2023

  59. [67]

    Google scanned objects: A high- quality dataset of 3d scanned household items

    Laura Downs, Anthony Francis, Nate Koenig, Brandon Kin- man, Ryan Hickman, Krista Reymann, Thomas B McHugh, and Vincent Vanhoucke. Google scanned objects: A high- quality dataset of 3d scanned household items. In2022 In- ternational Conference on Robotics and Automation (ICRA)...

    2022

  60. [68]

    Gans trained by a two time-scale update rule converge to a local nash equilib- rium.Advances in neural information processing systems, 30, 2017

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilib- rium.Advances in neural information processing systems, 30, 2017. 6

    2017

  61. [69]

    Demystifying mmd gans.arXiv preprint arXiv:1801.01401, 2018

    Mikołaj Bi ´nkowski, Danica J Sutherland, Michael Arbel, and Arthur Gretton. Demystifying mmd gans.arXiv preprint arXiv:1801.01401, 2018. 6

    2018 arXiv

  62. [70]

    Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980,

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980,

    arXiv

  63. [2014]

    A cylindrical cosmetic bag with a flora pattern and a zipper closure

    14 11 mesh result mesh result “A cylindrical cosmetic bag with a flora pattern and a zipper closure” “A bipedal theropod dinosaur, resembling a Tyrannosaurus Rex” “A colorful cylindrical pencil case with a pattern of geometric shapes and a zipper closure” “A Don’t Wake Daddy b...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.