Pith. sign in

REVIEW 4 major objections 6 minor 1 cited by

TRACE claims that anchoring an autoregressive video-diffusion model to explicit 3D mesh projections lets a single pipeline perform part-level geometric edits on 3D Gaussian scenes while keeping the rest of the scene stable.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-03 02:25 UTC pith:OG3FBWV6

load-bearing objection Sensible system-level idea with plausible qualitative results, but the evaluation never measures the structural-integrity claim it sells, and the missing appendix and unreleased artifacts make the key mechanism unverifiable. the 4 major comments →

arxiv 2604.01207 v2 pith:OG3FBWV6 submitted 2026-04-01 cs.CV

TRACE: High-Fidelity 3D Scene Editing via Tangible Reconstruction and Geometry-Aligned Contextual Video Masking

Jiyuan Hu , Zechuan Zhang , Zongxin Yang , Yi Yang This is my paper
classification cs.CV
keywords 3D Gaussian Splatting editingvideo diffusiongeometric anchoringmesh alignmentmulti-view consistencyobject insertionpart-level manipulationMV-TRACE dataset
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

TRACE tries to establish that explicit 3D geometry and generative video diffusion can be combined into one automated pipeline for editing 3D Gaussian Splatting scenes, including structural changes like shifting a statue's leg or replacing a component—something appearance-only editing methods cannot do. The paper argues that previous approaches either lack 3D grounding and collapse under multi-view viewing, or rely on explicit geometry that is slow, misaligned, and inharmonious with scene lighting. TRACE's answer is to treat meshes as geometric anchors that guide a video diffusion model, with three modules: multi-view anchor synthesis, two-phase mesh-scene registration, and contextually masked video inpainting. If the claim holds, a single 10-minute edit can reshape geometry and harmonize appearance without per-scene optimization.

Core claim

The central claim is that decoupling geometric anchoring from appearance harmonization resolves the trade-off between structural integrity and scene-consistent appearance in 3DGS editing. TRACE first generates multi-view consistent editing references with a 3D-LoRA fine-tuned sparse-view editor trained on the new MV-TRACE dataset, then registers a mesh to the scene through a two-phase coarse-to-fine alignment (Procrustes-style rotation locking followed by differentiable refinement with SDF penalization), and finally repaints the mesh projections with an autoregressive video diffusion model conditioned on gray-prefilled context and overlapping anchor latents. The paper reports that this pipel

What carries the argument

The load-bearing mechanism is Contextual Video Masking (CVM), which projects the aligned 3D mesh into video frames, extends raw masks to cover shadow and reflection zones, and feeds an autoregressive video diffusion model a gray-prefilled context plus overlapping anchor latents from the previous segment (Eq. 4: z*_j = Φ(z_j | y, M_j, V_j^gray, z*_{j-1}[1:k])). This is what supposedly guarantees 'strict long-range geometry consistency' during repainting, so the edited frames can be back-projected into a unified 3D Gaussian representation. The preceding modules—MV-TRACE-trained 3D-LoRA for anchor synthesis and two-phase TGA registration—exist to supply reliable, spatially synchronized 3D ancho

Load-bearing premise

The video-generation stage is assumed to keep geometry consistent across long camera moves when given only gray-filled context frames and a few overlapping frames from the previous video segment; the paper does not measure 3D geometry directly at this stage, so a drift there would silently corrupt the final 3D scene.

What would settle it

Render a scene with a large camera sweep around an inserted mesh, then compare the back-projected 3D Gaussian positions of the edited region against the ground-truth aligned mesh; if the mean positional error grows monotonically with angular distance from the anchor frames beyond a small threshold, the claimed long-range consistency of CVM is disproven. A simpler proxy: compute 3D IoU between the mesh projection and the final reconstructed splats for a 180-degree trajectory; a value near the 0.792 reported for TGA but low for CVM output would isolate the failure.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Part-level manipulation becomes feasible in a single automated pipeline: local pose shifts (e.g., raising a statue's leg) and component replacement on 3DGS scenes, which existing methods cannot do.
  • Edits complete in about 10 minutes on one GPU, matching feed-forward efficiency and avoiding the per-scene optimization of iterative baselines.
  • The MV-TRACE dataset is claimed to be the first multi-view consistent dataset for scene-coherent object addition and modification, providing a training resource for future 3D-aware editors.
  • The two-phase TGA alignment (rotation locking then differentiable refinement) lifts 2D IoU from 0.538 to 0.846 and 3D IoU to 0.792, suggesting the alignment recipe is transferable to other mesh-to-scene tasks.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper: the same 'explicit mesh as anchor + video diffusion for appearance' decomposition could apply to 4D/dynamic scenes, where the video prior already provides temporal consistency and the mesh anchor supplies a stable geometry skeleton.
  • Beyond the paper: the MV-TRACE dataset, described as 100k pairs of scene-coherent additions and modifications, could be reused as a benchmark for evaluating any multi-view editing method, not just TRACE itself.
  • Beyond the paper: a testable extension is to replace the private 3D-LoRA fine-tune with a stronger off-the-shelf multi-view editing model; if CVM still enforces geometry, the framework's generality is confirmed, whereas if IoU collapses, the claim narrows to the specific fine-tuned editor.
  • Beyond the paper: the 2D-IoU/3D-IoU metrics used in the TGA ablation could serve as a direct continuous measure of 'structural integrity' for future editing methods, supplementing the image-level DINO similarity.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. TRACE is a three-stage 3D Gaussian Splatting (3DGS) editing framework. First, Multi-view 3D-Anchor Synthesis fine-tunes a 3D-LoRA on a new private synthetic dataset (MV-TRACE) to generate geometrically consistent edited reference views. Second, Tangible Geometry Alignment (TGA) registers a mesh (from an LRM or generator) to the 3DGS scene via a two-phase Procrustes + differentiable SDF-refinement procedure. Third, Contextual Video Masking (CVM) repaints mesh projections in an autoregressive video diffusion model, and the generated video is back-projected into 3DGS. The paper reports qualitative results on eight held-out scenes and a quantitative comparison (Table 2) against six baselines, claiming large gains in CLIP directional similarity, DINO consistency, and aesthetic score with ~10-minute runtime, and it argues that TRACE uniquely supports part-level geometric manipulation while preserving structural integrity.

Significance. The proposed combination of explicit mesh anchoring with video diffusion is a plausible and timely direction for 3DGS editing, and the reported efficiency (~10 min per edit) is attractive compared with optimization-heavy baselines. The paper is also explicit about the failure modes of existing methods and provides extensive qualitative illustrations. However, the central claim—that TRACE preserves structural integrity while enabling part-level geometry edits—is not directly measured. The main evidence is an aggregate table over 8 scenes with no error bars or per-case breakdown, and the only geometry-related metrics (mask IoU) are computed against masks produced by the same pipeline. The private MV-TRACE dataset and the missing CVM appendix further limit reproducibility. The idea is promising and the engineering is substantial, but the current validation does not yet establish the paper's headline capability.

major comments (4)
  1. [Sec. 3.3, Eq. (4)] The paper claims that CVM 'ensures strict long-range geometry consistency' (Sec. 1) and enables 'physically-grounded rendering' (Sec. 3.3), but Eq. (4) conditions each autoregressive segment only on text, mask, gray-prefilled context, and overlapping latents. There is no depth term, reprojection constraint, or 3D geometric loss in the video-diffusion step. Table 2's DINO similarity measures global identity/consistency, not geometric alignment of the edited region. Section 4.3 explicitly defers the CVM ablation to an appendix that is not present in the manuscript. This is a load-bearing evidential gap for the 'structural integrity' claim: the mechanism that supposedly delivers the key capability is neither constrained nor measured at the stage where it matters.
  2. [Sec. 4.1, Table 2] The quantitative evaluation reports only aggregate means over 8 scenes and 48 editing cases, with no error bars, per-scene table, or significance tests. The reported CLIPdir improvement (0.1514 vs. 0.1013, +49.5%) could be driven by a few favorable scenes. Moreover, DINO similarity is a global feature consistency metric; it does not isolate whether the edited region is geometrically consistent across views. The paper should provide per-case results, variance statistics, and a direct geometric metric for the edited region (e.g., multi-view reprojection error, depth consistency, or chamfer distance against the mesh anchor) before the structural-integrity claim can be accepted.
  3. [Sec. 3.1, Eq. (1) and Sec. 4.3, Tables 3-4] The evaluation partly reuses the machinery used to train the system. Semantic alignment is measured with CLIP directional similarity, while the training loss L_tex in Eq. (1) is itself a text-alignment loss on a CLIP-family model; the mask IoU metrics in Tables 3 and 4 are computed against masks that come from the same anchor/mesh projection pipeline, not an independent ground truth. The paper should clarify how IoUalign and IoU2D/IoU3D are defined and, ideally, add evaluation on held-out real scenes with manually annotated or externally generated masks to break the circularity.
  4. [Sec. 3.1, MV-TRACE dataset] The method's first stage depends on fine-tuning a 3D-LoRA on the private MV-TRACE dataset. The dataset is not released, and the description (dense spherical sampling, 'human-assisted spatial alignment', view-pair filtering) lacks quantitative statistics (number of scenes/assets, view distribution, editing types, annotation protocol). Because the dataset is synthetic and the editing prior is fine-tuned on it, the paper should demonstrate that the learned prior transfers to real scenes beyond the eight shown, and ideally release the dataset or a representative subset to enable reproduction and independent verification.
minor comments (6)
  1. [Sec. 3.3] The parameters governing CVM, such as the angular sampling density threshold ρ and overlap anchor length k in Eq. (4), are never quantified. Without these values, the trajectory synthesis and autoregressive overlap strategy cannot be reproduced or compared.
  2. [Sec. 3.2, Eq. (2) and Eq. (3)] Notation is under-specified: P_i^Mono and P_i^Mesh are not defined precisely, and the indicator 1_SDF in Eq. (3) is introduced without explaining its values. The 'monocular depth priors' used for phase 1 are not identified (the text points to Sec. 4.3, but that section is not present in the manuscript).
  3. [Sec. 4.3, Tables 3-4] The definitions of LPIPSbg, CLIPdir, IoUalign, IoU2D, and IoU3D are not given. In particular, IoUalign in Table 3 is not the same as the IoU in the training loss (Eq. 1) unless a separate ground-truth mask set is used; the text should clarify.
  4. [Sec. 3.1] The exact composition of L_tex, L_cons, and LIoU is not specified (e.g., which CLIP variant, which perceptual loss, how IoU is computed). The hyperparameters λ1 and λ2 in Eq. (1) are also not reported.
  5. [Figures] Figure 3 and Figure 6 are referenced with precise sub-panels (e.g., 'Fig. 3 part c. leftmost image', 'Fig. 6, right column'), but some of these callouts are ambiguous without panel labels. Also, the qualitative figures are small; high-resolution crops would help verify the claimed structural integrity.
  6. [References] Several references are listed as arXiv preprints with 2026 dates (e.g., [68], [121]); if these are not yet publicly available, the authors should double-check the citations. Also, reference [6] is a duplicate of [5].

Circularity Check

1 steps flagged

TGA alignment IoU is reported against the same mask objective it optimizes; the main external benchmark remains independent.

specific steps
  1. fitted input called prediction [Sec. 3.2 (Eq. 3) and Sec. 4.3 (Table 4)]
    "These priors are aligned with the original scene’s geometry using masks from multiviews to determine the precise editing region. … We align the mesh against a sparse-view point cloud S3DGS (derived from the scene or feed-forward reconstruction) using a composite objective: Lrefine = λgeoLgeo + λmaskLmask + 1SDF·(λsdfLSDF + Rreg). … We evaluate alignment using Avg-2D IoU (projection overlap) and 3D IoU (volumetric overlap)."

    The TGA optimization (Eq. 3) includes an explicit λmaskLmask term, and the multiview masks used to define the editing region are the same pipeline-generated masks from Stage-1 anchor synthesis. Table 4 then offers Avg-2D IoU / projection overlap as the evidence that TGA achieves precise alignment. Unless those IoU values are computed against an independent ground-truth mask—which the paper never states—the reported metric is the fitting objective itself, so the alignment number is a fit report rather than a prediction. The 3D IoU is similarly tied to the S3DGS point cloud used in Eq. 3.

full rationale

The main benchmark comparison (Table 2) compares TRACE against six externally published baselines on eight held-out scenes from IN2N, BlendedMVS, and Mip-NeRF 360, using standard CLIPdir, DINO, and aesthetic scores. That part is not circular: the evaluation data and metrics are outside the training machinery. Eq. 1 trains the 3D anchor model on the authors' synthetic MV-TRACE dataset with textual, consistency, and mask-IoU losses; if IoUalign in Table 3 is computed on held-out data, this is ordinary generalization, though the paper does not specify the split. The clearest circular-adjacent step is the TGA ablation in Table 4, where the reported Avg-2D IoU and 3D IoU appear to measure overlap against the same multiview masks and sparse point cloud used as fitting targets in Eq. 3, making the ablation's alignment numbers forced by construction. Separately, Sec. 4.3 states 'the CVM refinement related ablation studies are shown in Appendix and Fig. 6,' but no appendix is present in the manuscript; this is an omitted-proof/evidence gap for the claimed 'strict long-range geometry consistency,' not itself a circularity. Self-citations appear in related work but are not used as a uniqueness theorem or as the sole load-bearing justification of a central result. Overall, one ablation metric reduces to its fitting objective, while the central held-out comparison retains independent content.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 1 invented entities

The central claim rests on a private dataset, hand-set loss weights, off-the-shelf mesh generators, and an assumed capability of video diffusion to maintain geometry from masks. The most honest measure of what the paper adds is the system integration; the load-bearing data and model weights are not independently supplied.

free parameters (4)
  • λ1, λ2 in Eq. (1) = not reported
    Weights for consistency and mask-IoU losses in Multi-view 3D-Anchor Synthesis; hand-set in training.
  • λgeo, λmask, λsdf and R_reg in Eq. (3) = not reported
    Weights for geometry, mask, SDF and regularization losses in TGA; hand-set.
  • Angular sampling density threshold ρ = not reported
    Threshold on angular sampling density for adaptive SLERP view insertion in CVM; chosen by hand.
  • Overlap anchor length k in Eq. (4) = not reported
    Number of overlapping latent frames inherited by each autoregressive segment; chosen by hand.
axioms (5)
  • domain assumption The scene comes from an accurate COLMAP reconstruction.
    TRACE starts from an initial COLMAP reconstruction; errors in SfM propagate to all subsequent alignment and editing (Sec. 3 intro).
  • ad hoc to paper MV-TRACE synthetic data is representative of real-world scene editing.
    The 3D-LoRA is fine-tuned on a private human-curated synthetic dataset; no independent validation that this distribution transfers to IN2N/BlendedMVS/Mip-NeRF scenes (Sec. 3.1).
  • domain assumption Off-the-shelf mesh generators produce meshes with sufficient fidelity for the desired edit.
    TGA's alignment and the final geometry are bounded by the quality of meshes from Hunyuan3D/Instant3dit/CraftMesh; poor proxies cause misalignment (Sec. 3.2).
  • domain assumption An autoregressive video diffusion model with mask/context conditioning preserves strict 3D geometry over long segments.
    CVM claims 'strict long-range geometry consistency' from gray-prefilled inputs and overlapping latents (Eq. 4), but no direct geometric metric is provided (Sec. 3.3).
  • domain assumption CLIP, DINO and aesthetic scores measure semantic alignment, multi-view consistency, and visual quality as claimed.
    Metrics are adopted from prior editing literature; they are appearance proxies, not structural/3D-consistency ground truth (Sec. 4.1).
invented entities (1)
  • MV-TRACE dataset no independent evidence
    purpose: Training and evaluation resource for the Multi-view 3D-Anchor Synthesis stage; claimed first multi-view consistent dataset for scene-coherent object addition/modification.
    No release link, no detailed statistics beyond 100k pairs/96 views, and no external use; the claim that it is 'first' cannot be independently checked.

pith-pipeline@v1.3.0-alltime-deepseek · 17079 in / 14779 out tokens · 138604 ms · 2026-08-03T02:25:13.624583+00:00 · methodology

0 comments
read the original abstract

Existing 3D Gaussian Splatting (3DGS) editing methods primarily focus on appearance modification and often struggle to support flexible geometry editing while preserving structural integrity and scene-consistent appearance. To address this limitation, we present TRACE, a mesh-guided 3DGS editing framework that automatically aligns explicit 3D geometry with Gaussian scenes and decouples Geometric Anchoring from Appearance Harmonization. First, Multi-view 3D-Anchor Synthesis, trained on our MV-TRACE dataset for scene-coherent object addition and modification, generates geometrically aligned editing anchors, while Tangible Geometry Alignment (TGA) performs coarse-to-fine mesh-scene registration. Then, Contextual Video Masking (CVM) integrates projected 3D anchors into an autoregressive video diffusion pipeline, harmonizing their appearance with the surrounding scene while maintaining multi-view consistency. We evaluate TRACE on eight held-out scenes across six editing categories. TRACE completes each edit in approximately 10 minutes on a single NVIDIA RTX Pro 6000 GPU. Extensive experiments demonstrate consistent improvements over existing methods in editing versatility, structural integrity, semantic alignment, multi-view consistency, and visual quality.

Figures

Figures reproduced from arXiv: 2604.01207 by Jiyuan Hu, Yi Yang, Zechuan Zhang, Zongxin Yang.

Figure 1
Figure 1. Figure 1: Representative showcases of TRACE. Top: Interactive edits on the “Stone Horse Statue” scene, demonstrating part-level geometric manipulation. Bottom: Di￾verse editing examples including object addition, texture synthesis, and style transfer. Abstract. We present TRACE, a mesh-guided 3DGS editing frame￾work that achieves automated, high-fidelity scene transformation. By anchoring video diffusion with explic… view at source ↗
Figure 2
Figure 2. Figure 2: Limitations of existing 3D editing approaches. (a) Geometric Insta [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Method overview. Our pipeline starts with (a) Multi-view 3D-Anchor Syn￾thesis, which generates geometrically-aligned reference views through 3D-LoRA and VLM guidance. These views are fed into (b) Tangible Geometry Alignment (TGA), where LRM-generated meshes are anchored in the scene via a two-stage alignment module. Subsequently, (c) Contextual Video Masking (CVM) propagates these edits across continuous c… view at source ↗
Figure 4
Figure 4. Figure 4: Overview of MV-TRACE dataset curation pipeline. (Top) [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Alignment pipeline. To edit the scene by adding sunglasses to the man’s face, the initial 3D asset is generated without orientation, resulting in severe misalignment and a reversed heading (see "Before Alignment"). Directly optimizing such a state would lead to a local optimum where the asset is oriented backwards. To resolve this, we employ a two-stage strategy as described in Sec. 3.2. The bottom row dem… view at source ↗
Figure 6
Figure 6. Figure 6: Visual Refinement and Trajectory Sampling in CVM. [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative comparisons of various scenes generated by different [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative comparison with direct video editing methods. [PITH_FULL_IMAGE:figures/full_fig_p013_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Multiview editing ablation. We compare editing across View 1 (red), View 2 (yellow), and View 3 (purple). No LoRA (top) lacks 3D awareness and tends to add objects in the same 2D position. Multi-Angle LoRA (middle) captures scene rota￾tion but suffers from background collapse and struggles to render local additions (e.g. lantern). In contrast, our 3D LoRA (bottom) ensures accurate and consistent object pla… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. VGGT-Edit: Feed-forward Native 3D Scene Editing with Residual Field Prediction

    cs.CV 2026-05 unverdicted novelty 7.0

    VGGT-Edit proposes a native 3D text-conditioned editing framework using depth-synchronized injection and residual field prediction, plus the DeltaScene dataset, outperforming 2D-lifting methods.

Reference graph

Works this paper leans on

120 extracted references · 36 linked inside Pith · cited by 1 Pith paper

  1. [1]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Avrahami, O., Patashnik, O., Fried, O., Nemchinov, E., Aberman, K., Lischinski, D., Cohen-Or, D.: Stable Flow: Vital Layers for Training-Free Image Editing. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 7877–7888 (2025) 4

    2025

  2. [2]

    In: Avidan, S., Brostow, G., Cissé, M., Farinella, G.M., Hassner, T

    Bar-Tal, O., Ofri-Amar, D., Fridman, R., Kasten, Y., Dekel, T.: Text2LIVE: Text- Driven Layered Image and Video Editing. In: Avidan, S., Brostow, G., Cissé, M., Farinella, G.M., Hassner, T. (eds.) Computer Vision – ECCV 2022. vol. 13675, pp. 707–723. Springer Nature Switzerland (2022).https://doi.org/10.1007/978- 3-031-19784-0_414

    doi:10.1007/978- 2022

  3. [3]

    Barda, A., Gadelha, M., Kim, V.G., Aigerman, N., Bermano, A.H., Groueix, T.: Instant3dit: Multiview inpainting for fast editing of 3d objects (2025) 5, 8

    2025

  4. [4]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Barron, J.T., Mildenhall, B., Verbin, D., Srinivasan, P.P., Hedman, P.: Mip- nerf 360: Unbounded anti-aliased neural radiance fields. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 5470– 5479 (2022) 10

    2022

  5. [6]

    arXiv preprint arXiv:2311.15127 (2023) 4

    Blattmann, A., Dockhorn, T., Kulal, S., Mendelevitch, D., Kilian, M., Lorenz, D., Levi, Y., English, Z., Voleti, V., Letts, A., et al.: Stable video diffusion: Scaling latent video diffusion models to large datasets. arXiv preprint arXiv:2311.15127 (2023) 4

    Pith/arXiv arXiv 2023

  6. [7]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Brack, M., Friedrich, F., Kornmeier, K., Tsaban, L., Schramowski, P., Kersting, K., Passos, A.: LEDITS++: Limitless image editing using text-to-image models. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 8861–8870 (2024) 4

    2024

  7. [8]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Brooks, T., Holynski, A., Efros, A.A.: Instructpix2pix: Learning to follow image editing instructions. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 18392–18402 (2023) 2, 4, 5

    2023

  8. [9]

    Blog Post (2024) 4

    Brooks, T., Peebles, B., Holmes, C., DePue, W., Guo, Y., Jing, L., Schnurr, D., Taylor, J., Luhman, T., Luhman, E., Ng, C., Wang, R., Ramesh, A.: Video generation models as world simulators. Blog Post (2024) 4

    2024

  9. [10]

    arXiv preprint arXiv:2505.22705 (2025).https://doi.org/10.48550/arXiv.2505.227054

    Cai, Q., Chen, J., Chen, Y., Li, Y., Long, F., Pan, Y., Qiu, Z., Zhang, Y., Gao, F., Xu, P., Wang, Y., Yu, K., Chen, W., Feng, Z., Gong, Z., Pan, J., Peng, Y., Tian, R., Wang, S., Zhao, B., Yao, T., Mei, T.: HiDream-I1: A High-Efficient Image Generative Foundation Model with Sparse Diffusion Transformer. arXiv preprint arXiv:2505.22705 (2025).https://doi....

    Pith/arXiv arXiv doi:10.48550/arxiv.2505.22705 2025

  10. [11]

    arXiv preprint arXiv:2510.23306 (2025) 5

    Chang, J., Ye, C., Wu, Y., Chen, Y., Zhang, Y., Luo, Z., Li, C., Zhi, Y., Han, X.: Reconviagen: Towards accurate multi-view 3d object reconstruction via genera- tion. arXiv preprint arXiv:2510.23306 (2025) 5

    arXiv 2025

  11. [12]

    arXiv preprint arXiv:2510.15857 (2025) 4

    Chen, J., Xue, L., Xu, Z., Pan, X., Yang, S., Qin, C., Yan, A., Zhou, H., Chen, Z., Huang, L., Zhou, T., Li, J., Savarese, S., Xiong, C., Xu, R.: Blip3o-next: Next frontier of native image generation. arXiv preprint arXiv:2510.15857 (2025) 4

    arXiv 2025

  12. [13]

    arXiv preprint arXiv:2511.23172 (2025) 2, 11, 12

    Chen, L., Li, R., Zhang, G., Wang, P., Zhang, L.: Fast multi-view consistent 3d editing with video priors. arXiv preprint arXiv:2511.23172 (2025) 2, 11, 12

    arXiv 2025

  13. [14]

    In: European Conference on Computer Vision

    Chen, M., Laina, I., Vedaldi, A.: Dge: Direct gaussian 3d editing by consistent multi-view editing. In: European Conference on Computer Vision. pp. 74–92. Springer (2024) 2, 5, 11, 12 16 J. Hu, Z. Zhang et al

    2024

  14. [15]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Chen, M., Xie, J., Laina, I., Vedaldi, A.: Shap-editor: Instruction-guided latent 3d editing in seconds. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 26456–26466 (2024) 5

    2024

  15. [16]

    arXiv preprint arXiv:2311.14521 (2023) 2, 5, 11, 12

    Chen, Y., Chen, Z., Zhang, C., Wang, F., Yang, X., Wang, Y., et al.: Gaussianed- itor: Swift and controllable 3d editing with gaussian splatting. arXiv preprint arXiv:2311.14521 (2023) 2, 5, 11, 12

    Pith/arXiv arXiv 2023

  16. [17]

    In: The Eleventh International Con- ference on Learning Representations (2023) 4

    Couairon, G., Verbeek, J., Schwenk, H., Cord, M.: DiffEdit: Diffusion-based se- mantic image editing with mask guidance. In: The Eleventh International Con- ference on Learning Representations (2023) 4

    2023

  17. [18]

    arXiv preprint arXiv:2510.26583 (2025)

    Cui, Y., Chen, H., Deng, H., Huang, X., Li, X., Liu, J., Liu, Y., Luo, Z., Wang, J., Wang, W., Wang, Y., Wang, C., Zhang, F., Zhao, Y., Pan, T., Li, X., Hao, Z., Ma, W., Chen, Z., Ao, Y., Huang, T., Wang, Z., Wang, X.: Emu3.5: Native Multimodal Models are World Learners. arXiv preprint arXiv:2510.26583 (2025). https://doi.org/10.48550/arXiv.2510.265834

    Pith/arXiv arXiv doi:10.48550/arxiv.2510.26583 2025

  18. [19]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Decatur, D., Lang, I., Aberman, K., Hanocka, R.: 3d paintbrush: Local stylization of 3d shapes with cascaded score distillation. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 4473–4483 (2024) 5

    2024

  19. [20]

    arXiv preprint arXiv:2307.05663 (2023) 6

    Deitke, M., Liu, R., Wallingford, M., Ngo, H., Michel, O., Kusupati, A., Fan, A., Laforte, C., Voleti, V., Gadre, S.Y., VanderBilt, E., Kembhavi, A., Vondrick, C., Gkioxari, G., Ehsani, K., Schmidt, L., Farhadi, A.: Objaverse-xl: A universe of 10m+ 3d objects. arXiv preprint arXiv:2307.05663 (2023) 6

    Pith/arXiv arXiv 2023

  20. [21]

    arXiv preprint arXiv:2505.14683 (2025).https://doi.org/10.48550/ arXiv.2505.146834

    Deng, C., Zhu, D., Li, K., Gou, C., Li, F., Wang, Z., Zhong, S., Yu, W., Nie, X., Song, Z., Shi, G., Fan, H.: Emerging Properties in Unified Multimodal Pre- training. arXiv preprint arXiv:2505.14683 (2025).https://doi.org/10.48550/ arXiv.2505.146834

    Pith/arXiv arXiv doi:10.48550/arxiv.2505.14683 2025

  21. [22]

    arXiv preprint arXiv:2509.24979 (2025).https: //doi.org/10.48550/arXiv.2509.249794

    Dong, H., Wang, W., Li, C., Lin, D.: Wan-Alpha: High-Quality Text-to-Video Generation with Alpha Channel. arXiv preprint arXiv:2509.24979 (2025).https: //doi.org/10.48550/arXiv.2509.249794

    doi:10.48550/arxiv.2509.24979 2025

  22. [23]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Dong, S., Ding, L., Huang, Z., Wang, Z., Xue, T., Xu, D.: Interactive3d: Create what you want by interactive 3d generation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 4999–5008 (2024) 5

    2024

  23. [24]

    In: Forty-First International Conference on Machine Learning (2024) 4

    Esser, P., Kulal, S., Blattmann, A., Entezari, R., Müller, J., Saini, H., Levi, Y., Lorenz, D., Sauer, A., Boesel, F., Podell, D., Dockhorn, T., English, Z., Rombach, R.: Scaling Rectified Flow Transformers for High-Resolution Image Synthesis. In: Forty-First International Conference on Machine Learning (2024) 4

    2024

  24. [25]

    fal: Qwen-image-edit-2511-multiple-angles-lora.https://huggingface.co/fal/ Qwen-Image-Edit-2511-Multiple-Angles-LoRA(2025) 12

    2025

  25. [26]

    In: SIGGRAPH Asia 2024 Conference Papers

    Fujiwara, H., Mukuta, Y., Harada, T.: Style-nerf2nerf: 3d style transfer from style-aligned multi-view images. In: SIGGRAPH Asia 2024 Conference Papers. pp. 1–10 (2024) 5

    2024

  26. [27]

    arXiv preprint arxiv:2307.10373 (2023) 4

    Geyer, M., Bar-Tal, O., Bagon, S., Dekel, T.: Tokenflow: Consistent diffusion features for consistent video editing. arXiv preprint arxiv:2307.10373 (2023) 4

    Pith/arXiv arXiv 2023

  27. [28]

    Gomel, E., Wolf, L.: Diffusion-based attention warping for consistent 3d scene editing (2024),https://arxiv.org/abs/2412.079845

    Pith/arXiv arXiv 2024

  28. [29]

    Blog Post (2024) 4

    Google DeepMind: Veo 2: Generating high quality videos with improved under- standing. Blog Post (2024) 4

    2024

  29. [30]

    Blog Post (2024) 4

    Google DeepMind: Veo: High-quality video generation from text prompts. Blog Post (2024) 4

    2024

  30. [31]

    Blog Post (2025) 4 TRACE 17

    Google DeepMind: Veo 3: Video generation with synchronized audio. Blog Post (2025) 4 TRACE 17

    2025

  31. [32]

    arXiv preprint arXiv:2312.02087 (2023) 4

    Gu,Y.,Zhou,Y.,Wu,B.,Yu,L.,Liu,J.W.,Zhao,R.,Wu,J.Z.,Zhang,D.J.,Shou, M.Z., Tang, K.: Videoswap: Customized video subject swapping with interactive semantic point correspondence. arXiv preprint arXiv:2312.02087 (2023) 4

    Pith/arXiv arXiv 2023

  32. [33]

    0: A larger-scale dataset of multi-view images

    Han, X., Wu, Y., Shi, L., Liu, H., Liao, H., Qiu, L., Yuan, W., Gu, X., Dong, Z., Cui, S.: Mvimgnet2. 0: A larger-scale dataset of multi-view images. arXiv preprint arXiv:2412.01430 (2024) 6

    Pith/arXiv arXiv 2024

  33. [34]

    In: Proceedings of the IEEE/CVF Interna- tional Conference on Computer Vision (2023) 5, 10

    Haque, A., Tancik, M., Efros, A., Holynski, A., Kanazawa, A.: Instruct-nerf2nerf: Editing 3d scenes with instructions. In: Proceedings of the IEEE/CVF Interna- tional Conference on Computer Vision (2023) 5, 10

    2023

  34. [35]

    Hertz, A., Mokady, R., Tenenbaum, J., Aberman, K., Pritch, Y., Cohen-or, D.: Prompt-to-PromptImageEditingwithCross-AttentionControl.In:TheEleventh International Conference on Learning Representations (2023) 4

    2023

  35. [36]

    arXiv preprint arXiv:2106.09685 (2021) 4

    Hu, E.J., Shen, Y., Wallis, P., Allen-Zhu, Z., Li, Y., Wang, S., Wang, L., Chen, W.: Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685 (2021) 4

    Pith/arXiv arXiv 2021

  36. [37]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Huang, Y., Xie, L., Wang, X., Yuan, Z., Cun, X., Ge, Y., Zhou, J., Dong, C., Huang, R., Zhang, R., Shan, Y.: SmartEdit: Exploring complex instruction-based image editing with multimodal large language models. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 8362–8371 (2024) 4

    2024

  37. [38]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) (2025) 2, 4

    Jiang, Z., Han, Z., Mao, C., Zhang, J., Pan, Y., Liu, Y.: Vace: All-in-one video creation and editing. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) (2025) 2, 4

    2025

  38. [39]

    Jincheng, J., Wu, Y., Cai, Y., Liu, L.: Craftmesh: High-fidelity generative mesh manipulation via poisson seamless fusion (2025),https://arxiv.org/abs/2509. 136888

    2025

  39. [40]

    In: ICLR (2024) 4

    Ju, X., Zeng, A., Bian, Y., Liu, S., Xu, Q.: PnP Inversion: Boosting Diffusion- based Editing with 3 Lines of Code. In: ICLR (2024) 4

    2024

  40. [41]

    Kawar,B.,Zada,S.,Lang,O.,Tov,O.,Chang,H.,Dekel,T.,Mosseri,I.,Irani,M.: Imagic:Text-basedrealimageeditingwithdiffusionmodels.In:Proceedingsofthe IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 6007–6017 (2023) 4

    2023

  41. [42]

    ACM Transactions on Graphics42(4) (July 2023),https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/2

    Kerbl, B., Kopanas, G., Leimkühler, T., Drettakis, G.: 3d gaussian splatting for real-time radiance field rendering. ACM Transactions on Graphics42(4) (July 2023),https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/2

    2023

  42. [43]

    Khalid, U., Iqbal, H., Farooq, A., Hua, J., Chen, C.: 3dego: 3d editing on the go! (2024) 2

    2024

  43. [44]

    arXiv preprint arXiv:2407.11394 (2024) 5

    Kim, J., Lee, S., Shin, J., Choi, J., Shim, H.: Dreamcatalyst: Fast and high-quality 3d editing via controlling editability and identity preservation. arXiv preprint arXiv:2407.11394 (2024) 5

    Pith/arXiv arXiv 2024

  44. [45]

    arXiv preprint arXiv:2412.03603 (2025).https://doi.org/10.48550/arXiv.2412.036034

    Kong,W.,Tian,Q.,Zhang,Z.,Min,R.,Dai,Z.,Zhou,J.,Xiong,J.,Li,X.,Wu,B., Zhang, J., Wu, K., Lin, Q., Yuan, J., Long, Y., Wang, A., Wang, A., Li, C., Huang, D.,Yang,F.,Tan,H.,Wang,H.,Song,J.,Bai,J.,Wu,J.,Xue,J.,Wang,J.,Wang, K., Liu, M., Li, P., Li, S., Wang, W., Yu, W., Deng, X., Li, Y., Chen, Y., Cui, Y., Peng, Y., Yu, Z., He, Z., Xu, Z., Zhou, Z., Xu, Z., T...

    Pith/arXiv arXiv doi:10.48550/arxiv.2412.03603 2025

  45. [46]

    arXiv preprint arXiv:2403.14468 (2024) 4 18 J

    Ku, M., Wei, C., Ren, W., Yang, H., Chen, W.: Anyv2v: A tuning-free framework for any video-to-video editing tasks. arXiv preprint arXiv:2403.14468 (2024) 4 18 J. Hu, Z. Zhang et al

    Pith/arXiv arXiv 2024

  46. [47]

    Blog Post (2024) 12

    Kuaishou: Kling ai. Blog Post (2024) 12

    2024

  47. [48]

    In: Proceed- ings of the IEEE/CVF International Conference on Computer Vision

    Kulikov, V., Kleiner, M., Huberman-Spiegelglas, I., Michaeli, T.: FlowEdit: Inversion-Free Text-Based Editing Using Pre-Trained Flow Models. In: Proceed- ings of the IEEE/CVF International Conference on Computer Vision. pp. 19721– 19730 (2025) 4

    2025

  48. [49]

    Labs, B.F.: Flux.https://github.com/black-forest-labs/flux(2024) 2

    2024

  49. [50]

    Blog Post (2024) 4

    Labs, B.F.: Flux. Blog Post (2024) 4

    2024

  50. [51]

    Labs, B.F., Batifol, S., Blattmann, A., Boesel, F., Consul, S., Diagne, C., Dock- horn, T., English, J., English, Z., Esser, P., Kulal, S., Lacey, K., Levi, Y., Li, C., Lorenz, D., Müller, J., Podell, D., Rombach, R., Saini, H., Sauer, A., Smith, L.: Flux.1 kontext: Flow matching for in-context image generation and editing in latent space (2025),https://a...

    Pith/arXiv arXiv 2025

  51. [52]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Lee, D.I., Park, H., Seo, J., Park, E., Park, H., Baek, H.D., Shin, S., Kim, S., Kim, S.: Editsplat: Multi-view fusion and attention-guided optimization for view- consistent 3d scene editing with 3d gaussian splatting. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 11135–11145 (2025) 10, 11, 12

    2025

  52. [53]

    arXiv preprint arXiv:2505.01322 (2025) 2, 5

    Li, C., Wang, W., Li, Q., Lepri, B., Sebe, N., Nie, W.: Freeinsert: Disentangled text-guided object insertion in 3d gaussian scene without spatial priors. arXiv preprint arXiv:2505.01322 (2025) 2, 5

    arXiv 2025

  53. [54]

    arXiv preprint arXiv:2510.16888 (2025) 4

    Li, Z., Liu, Z., Zhang, Q., Lin, B., Yuan, S., Yan, Z., Ye, Y., Yu, W., Niu, Y., Yuan, L.: Uniworld-v2: Reinforce image editing with diffusion negative-aware finetuning and mllm implicit feedback. arXiv preprint arXiv:2510.16888 (2025) 4

    Pith/arXiv arXiv 2025

  54. [55]

    Lin, B., Li, Z., Cheng, X., Niu, Y., Ye, Y., He, X., Yuan, S., Yu, W., Wang, S., Ge, Y., Pang, Y., Yuan, L.: UniWorld-V1: High-Resolution Semantic Encoders for Unified Visual Understanding and Generation (Jun 2025).https://doi.org/ 10.48550/arXiv.2506.031474

    Pith/arXiv arXiv doi:10.48550/arxiv.2506.03147 2025

  55. [56]

    arXiv preprint arXiv:2511.10647 (2025) 9

    Lin, H., Chen, S., Liew, J.H., Chen, D.Y., Li, Z., Shi, G., Feng, J., Kang, B.: Depth anything 3: Recovering the visual space from any views. arXiv preprint arXiv:2511.10647 (2025) 9

    Pith/arXiv arXiv 2025

  56. [57]

    arXiv preprint arXiv:2504.17761 (2025) 4

    Liu, S., Han, Y., Xing, P., Yin, F., Wang, R., Cheng, W., Liao, J., Wang, Y., Fu, H., Han, C., et al.: Step1x-edit: A practical framework for general image editing. arXiv preprint arXiv:2504.17761 (2025) 4

    Pith/arXiv arXiv 2025

  57. [58]

    In: ICASSP 2025 - 2025 IEEE In- ternational Conference on Acoustics, Speech and Signal Processing (ICASSP)

    Lu, L., Li, R., Zhao, Y., Zhang, X., Wei, H., Li, R.: Mmeditor: Multimodal prompt-driven 3d gaussian splatting editing. In: ICASSP 2025 - 2025 IEEE In- ternational Conference on Acoustics, Speech and Signal Processing (ICASSP). pp. 1–5 (2025).https://doi.org/10.1109/ICASSP49660.2025.108877792

    arXiv 2025

  58. [59]

    In: International Conference on Learning Representations (2022) 4

    Meng, C., He, Y., Song, Y., Song, J., Wu, J., Zhu, J.Y., Ermon, S.: SDEdit: Guided Image Synthesis and Editing with Stochastic Differential Equations. In: International Conference on Learning Representations (2022) 4

    2022

  59. [60]

    Miao, B., Wei, R., Ge, Z., sun, X., Gao, S., Zhu, J., Wang, R., Tang, S., Xiao, J., Tang, R., Li, J.: Towards physically executable 3d gaussian for embodied navigation (2025),https://arxiv.org/abs/2510.213072

    arXiv 2025

  60. [61]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Mokady, R., Hertz, A., Aberman, K., Pritch, Y., Cohen-Or, D.: NULL-text inver- sion for editing real images using guided diffusion models. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 6038–6047 (2023) 4

    2023

  61. [62]

    arXiv preprint arXiv:2511.00062 (2025) 4 TRACE 19

    NVIDIA: World simulation with video foundation models for physical ai. arXiv preprint arXiv:2511.00062 (2025) 4 TRACE 19

    Pith/arXiv arXiv 2025

  62. [63]

    Oquab, M., Darcet, T., Moutakanni, T., Vo, H.V., Szafraniec, M., Khalidov, V., Fernandez, P., Haziza, D., Massa, F., El-Nouby, A., Howes, R., Huang, P.Y., Xu, H., Sharma, V., Li, S.W., Galuba, W., Rabbat, M., Assran, M., Ballas, N., Synnaeve, G., Misra, I., Jegou, H., Mairal, J., Labatut, P., Joulin, A., Bojanowski, P.: Dinov2: Learning robust visual feat...

    2023

  63. [64]

    In: ACM SIGGRAPH 2023 Conference Proceedings

    Parmar, G., Kumar Singh, K., Zhang, R., Li, Y., Lu, J., Zhu, J.Y.: Zero-shot Image-to-Image Translation. In: ACM SIGGRAPH 2023 Conference Proceedings. pp. 1–11. SIGGRAPH ’23, Association for Computing Machinery (2023).https: //doi.org/10.1145/3588432.35915134

    arXiv 2023

  64. [65]

    arXiv preprint arXiv:2503.09642 (2025).https://doi.org/10.48550/arXiv.2503.096424

    Peng, X., Zheng, Z., Shen, C., Young, T., Guo, X., Wang, B., Xu, H., Liu, H., Jiang, M., Li, W., Wang, Y., Ye, A., Ren, G., Ma, Q., Liang, W., Lian, X., Wu, X., Zhong, Y., Li, Z., Gong, C., Lei, G., Cheng, L., Zhang, L., Li, M., Zhang, R., Hu, S., Huang, S., Wang, X., Zhao, Y., Wang, Y., Wei, Z., You, Y.: Open-Sora 2.0: Training a Commercial-Level Video G...

    Pith/arXiv arXiv doi:10.48550/arxiv.2503.09642 2025

  65. [66]

    arXiv preprint arXiv:2209.14988 (2022) 2

    Poole, B., Jain, A., Barron, J.T., Mildenhall, B.: Dreamfusion: Text-to-3d using 2d diffusion. arXiv preprint arXiv:2209.14988 (2022) 2

    Pith/arXiv arXiv 2022

  66. [67]

    arXiv:2303.09535 (2023) 4

    Qi, C., Cun, X., Zhang, Y., Lei, C., Wang, X., Shan, Y., Chen, Q.: Fatezero: Fusing attentions for zero-shot text-based video editing. arXiv:2303.09535 (2023) 4

    Pith/arXiv arXiv 2023

  67. [68]

    Qin, H., Sun, Y., Wang, M., Kong, M., Lu, M., Zhu, Q.: Variation-aware flexible 3d gaussian editing (2026),https://arxiv.org/abs/2602.116385

    arXiv 2026

  68. [69]

    In: International conference on machine learning

    Radford,A.,Kim,J.W.,Hallacy,C.,Ramesh,A.,Goh,G.,Agarwal,S.,Sastry,G., Askell, A., Mishkin, P., Clark, J., et al.: Learning transferable visual models from natural language supervision. In: International conference on machine learning. pp. 8748–8763. PMLR (2021) 10

    2021

  69. [70]

    In: The Thirteenth International Conference on Learning Representations (2025) 4

    Rout, L., Chen, Y., Ruiz, N., Caramanis, C., Shakkottai, S., Chu, W.S.: Semantic Image Inversion and Editing using Rectified Stochastic Differential Equations. In: The Thirteenth International Conference on Learning Representations (2025) 4

    2025

  70. [71]

    Blog Post (2023) 4

    Runway: Gen-2: Generate novel videos with text, images or video clips. Blog Post (2023) 4

    2023

  71. [72]

    Blog Post (2024) 4

    Runway: Gen-3 alpha: A new frontier for video generation. Blog Post (2024) 4

    2024

  72. [73]

    arXiv preprint arXiv:2511.16624 (2025) 5

    SAM 3D Team, Chen, X., Chu, F.J., Gleize, P., Liang, K.J., Sax, A., et al.: Sam 3d: 3dfy anything in images. arXiv preprint arXiv:2511.16624 (2025) 5

    Pith/arXiv arXiv 2025

  73. [74]

    Schuhmann, C., Beaumont, R., Vencu, R., Gordon, C., Wightman, R., Cherti, M., Coombes, T., Katta, A., Mullis, C., Wortsman, M., Schramowski, P., Kundurthy, S., Crowson, K., Schmidt, L., Kaczmarczyk, R., Jitsev, J.: Laion-5b: An open large-scale dataset for training next generation image-text models (2022),https: //arxiv.org/abs/2210.0840211

    Pith/arXiv arXiv 2022

  74. [75]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Sella, E., Fiebelman, G., Hedman, P., Averbuch-Elor, H.: Vox-e: Text-guided voxel editing of 3d objects. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 430–440 (2023) 5

    2023

  75. [76]

    SIGGRAPH 2024 (2024) 4

    Shi, X., Huang, Z., Wang, F.Y., Bian, W., Li, D., Zhang, Y., Zhang, M., Cheung, K.C., See, S., Qin, H., et al.: Motion-i2v: Consistent and controllable image-to- video generation with explicit motion modeling. SIGGRAPH 2024 (2024) 4

    2024

  76. [77]

    In: Interna- tional Conference on Learning Representations (2021) 4

    Song, J., Meng, C., Ermon, S.: Denoising Diffusion Implicit Models. In: Interna- tional Conference on Learning Representations (2021) 4

    2021

  77. [78]

    In: Proceedings of the AAAI Con- ference on Artificial Intelligence

    Song, W., Jiang, H., Yang, Z., Cheng, Z., Quan, R., Yang, Y.: Insert anything: Image insertion via in-context editing in dit. In: Proceedings of the AAAI Con- ference on Artificial Intelligence. vol. 40, pp. 9097–9105 (2026) 4 20 J. Hu, Z. Zhang et al

    2026

  78. [79]

    Computer Graph- ics Forum43(2024),https://api.semanticscholar.org/CorpusID:273824919 2, 5

    Sun, Y., Tian, R., Han, X., Liu, X., Zhang, Y., Xu, K.: Gseditpro: 3d gaussian splatting editing with attention-based progressive localization. Computer Graph- ics Forum43(2024),https://api.semanticscholar.org/CorpusID:273824919 2, 5

    2024

  79. [80]

    Team, D.: Lucy edit: Open-weight text-guided video editing (2025) 12

    2025

  80. [81]

    Team, T.H.: Hunyuan3d 2.1: From images to high-fidelity 3d assets with production-ready pbr material (2025) 5, 8

    2025

Showing first 80 references.