pith. sign in

arxiv: 2606.04797 · v1 · pith:IQIMISDNnew · submitted 2026-06-03 · 💻 cs.CV · cs.LG

Crafting Your Evolving Dreams: Concept-Incremental Versatile Customization

Jiahua Dong , Wenqi Liang , Hongliu Li , Yang Cong , Duzhen Zhang , Hanbin Zhao , Henghui Ding , Yulun Zhang
show 2 more authors
Salman Khan Fahad Shahbaz Khan
This is my paper

Pith reviewed 2026-06-28 06:41 UTC · model grok-4.3

classification 💻 cs.CV cs.LG
keywords continual learningdiffusion modelsconcept customizationLoRAcatastrophic forgettingmulti-concept generationimage personalization
0
0 comments X

The pith

A diffusion model can incrementally learn new personalized concepts without forgetting earlier ones or neglecting details in multi-concept images.

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

The paper develops a Continually Customizable Diffusion Model (CCDM) that lets users add personalized concepts to a diffusion model one after another. Existing custom diffusion models treat the set of concepts as fixed and suffer from catastrophic forgetting of old concepts plus neglect of their details when new ones arrive. CCDM counters forgetting through an attribute-decoupled LoRA module that isolates each concept's attributes and a relevance-guided aggregation step that borrows useful correlations across tasks. A separate controllable regional context synthesis step ensures that multiple concepts can be composed in one image with clear region boundaries and no semantic bleed. If these mechanisms work, users would no longer need to retrain from scratch or accept degraded outputs every time their collection of desired concepts grows.

Core claim

The central claim is that an attribute-decoupled LoRA module together with relevance-guided aggregation preserves concept-specific attributes of each incremental task while exploiting beneficial inter-task correlations, and that a controllable regional context synthesis strategy produces multi-concept images with semantic independence between user-defined regions and smooth boundary transitions, thereby solving both catastrophic forgetting and concept neglect in continual customization of diffusion models.

What carries the argument

Attribute-decoupled LoRA (AD-LoRA) module, which separates concept attributes so that each task's unique features remain isolated while still permitting controlled aggregation across tasks.

If this is right

  • New customization tasks can be added without requiring full retraining or post-hoc fixes that degrade prior performance.
  • Multi-concept images maintain region-specific semantics and avoid attribute mixing at boundaries.
  • Inter-task relevance can be used to improve learning speed or quality of later tasks without harming earlier ones.
  • The model supports versatile user conditions for region placement during composition.

Where Pith is reading between the lines

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

  • The same modular separation of attributes could be tested on other parameter-efficient fine-tuning methods beyond LoRA.
  • If the approach scales, personal image generators might support lifelong user collections measured in dozens of concepts rather than a handful.
  • The regional synthesis component might generalize to video or 3D generation where temporal or spatial independence is also required.

Load-bearing premise

Decoupling attributes inside the LoRA updates will keep each concept's identity intact even when later tasks are learned and their parameters are aggregated.

What would settle it

Train CCDM sequentially on five unrelated concepts, then measure whether images of the first concept retain the same identity, detail fidelity, and prompt adherence as the single-task baseline.

Figures

Figures reproduced from arXiv: 2606.04797 by Duzhen Zhang, Fahad Shahbaz Khan, Hanbin Zhao, Henghui Ding, Hongliu Li, Jiahua Dong, Salman Khan, Wenqi Liang, Yang Cong, Yulun Zhang.

Figure 1
Figure 1. Figure 1: Illustration of the proposed CIVC problem. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Demonstration of our model’s scalability in supporting versatile concept customization tasks, including single/multi-concept synthesis, editing, [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Demonstration of (a) the attribute-decoupled LoRA (AD-LoRA) [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Demonstration of the controllable regional context synthesis [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Illustration of transforming a motion trajectory into bounding boxes. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Visualization of exemplary cases from 35 continuous concept [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative comparisons of single- and multi-concept text-to-image customization generated by SDXL [ [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative comparisons of single- and multi-concept text-to-image customization generated by FLUX.1 [ [PITH_FULL_IMAGE:figures/full_fig_p011_8.png] view at source ↗
Figure 10
Figure 10. Figure 10: Qualitative comparisons of style-transfer text-to-image customiza [PITH_FULL_IMAGE:figures/full_fig_p011_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Qualitative comparisons of single- and multi-concept text-to-video customization under the CIVC setting. [PITH_FULL_IMAGE:figures/full_fig_p012_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Qualitative comparisons of style-transfer text-to-video customiza [PITH_FULL_IMAGE:figures/full_fig_p012_12.png] view at source ↗
Figure 14
Figure 14. Figure 14: Qualitative comparisons of single- and multi-concept text-to-3D customization under the CIVC setting. [PITH_FULL_IMAGE:figures/full_fig_p013_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Qualitative ablation studies of single-concept text-to-image [PITH_FULL_IMAGE:figures/full_fig_p013_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Qualitative ablation studies of multi-concept text-to-image customization results generated by SDXL [ [PITH_FULL_IMAGE:figures/full_fig_p014_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Ablation studies of single-concept text-to-3D customization. [PITH_FULL_IMAGE:figures/full_fig_p014_17.png] view at source ↗
Figure 19
Figure 19. Figure 19: Qualitative results of multi-concept text-to-image customization generated by SDXL [ [PITH_FULL_IMAGE:figures/full_fig_p015_19.png] view at source ↗
read the original abstract

Custom diffusion models (CDMs) have garnered significant interest owing to their remarkable capacity for generating personalized concepts. However, the majority of CDMs unrealistically presume that the user's collection of personalized concepts is static and incapable of incremental growth over time. Furthermore, they exhibit significant catastrophic forgetting and concept neglect of previously learned concepts when incrementally learning a sequence of new ones. To resolve the above challenges, we develop a novel Continually Customizable Diffusion Model (CCDM), enabling users to perform concept-incremental versatile customization. Specifically, we design an attribute-decoupled LoRA (AD-LoRA) module and a relevance-guided AD-LoRA aggregation strategy to mitigate catastrophic forgetting. They can preserve concept-specific attributes of each task and leverage beneficial inter-task correlations to enhance the continual learning of new customization tasks. Additionally, to address the challenge of concept neglect, we propose a controllable regional context synthesis strategy that performs multi-concept composition in alignment with user-provided conditions. This strategy enhances the overall consistency in multi-concept synthesis by guaranteeing semantic independence between user-defined regions and their smooth boundary transitions. Experiments show our CCDM exhibits significant improvements over baseline methods.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The manuscript proposes a Continually Customizable Diffusion Model (CCDM) for concept-incremental versatile customization of diffusion models. It introduces an attribute-decoupled LoRA (AD-LoRA) module paired with a relevance-guided AD-LoRA aggregation strategy to mitigate catastrophic forgetting by preserving task-specific attributes while exploiting inter-task correlations, and a controllable regional context synthesis strategy to prevent concept neglect during multi-concept composition. The central claim is that these components together enable incremental learning of new personalized concepts without the forgetting and neglect observed in prior custom diffusion models, with experiments purportedly demonstrating significant improvements over baselines.

Significance. If the empirical results hold under rigorous evaluation, the work would be significant for continual and lifelong learning in generative models, as it targets practical limitations in evolving user-driven personalization. The introduction of named modules (AD-LoRA, relevance-guided aggregation, controllable regional synthesis) that aim to decouple attributes and enforce regional independence represents a targeted architectural response to known issues in incremental fine-tuning of diffusion models.

major comments (2)
  1. [Abstract] Abstract: the assertion that 'Experiments show our CCDM exhibits significant improvements over baseline methods' supplies no quantitative metrics, dataset names/sizes, baseline descriptions, or ablation results. This absence makes it impossible to determine whether the data support the central claim that AD-LoRA plus relevance-guided aggregation solves catastrophic forgetting without introducing new interference.
  2. [Abstract] The weakest assumption—that the attribute-decoupled LoRA module together with relevance-guided aggregation will preserve concept-specific attributes without requiring post-hoc adjustments—is presented without any derivation or control experiment showing that the relevance scores reduce to quantities independent of fitted parameters from prior tasks.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed feedback on our manuscript. We address each major comment below and will revise the abstract to provide greater specificity and clarity while preserving the manuscript's core contributions.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the assertion that 'Experiments show our CCDM exhibits significant improvements over baseline methods' supplies no quantitative metrics, dataset names/sizes, baseline descriptions, or ablation results. This absence makes it impossible to determine whether the data support the central claim that AD-LoRA plus relevance-guided aggregation solves catastrophic forgetting without introducing new interference.

    Authors: We agree that the abstract would be strengthened by including concrete details. In the revised manuscript we will expand the abstract to report key quantitative metrics (e.g., forgetting reduction percentages and multi-concept composition scores), the datasets used (including number of concepts and images per task), the specific baseline methods compared, and references to the ablation studies that isolate the contribution of AD-LoRA and relevance-guided aggregation. These elements already appear in Sections 4 and 5; moving concise versions into the abstract will directly address the concern about supporting the central claim. revision: yes

  2. Referee: [Abstract] The weakest assumption—that the attribute-decoupled LoRA module together with relevance-guided aggregation will preserve concept-specific attributes without requiring post-hoc adjustments—is presented without any derivation or control experiment showing that the relevance scores reduce to quantities independent of fitted parameters from prior tasks.

    Authors: The derivation of relevance scores and their claimed independence from prior-task parameters is provided in Section 3.2, where the attribute-decoupling formulation and the aggregation formula are shown to operate on per-task attribute embeddings. Nevertheless, we acknowledge that the abstract itself does not reference this derivation or any supporting control. We will revise the abstract to briefly note the independence property and will add a short control experiment (new panel in an existing ablation figure) that explicitly verifies relevance scores remain stable when prior-task parameters are frozen. This addition will be included in the next version. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper introduces named modules (AD-LoRA, relevance-guided aggregation, controllable regional context synthesis) as design choices to address forgetting and neglect, then reports empirical improvements over baselines. No equations, parameter fits, or self-citation chains are shown that reduce any claimed prediction or uniqueness result to the inputs by construction. The central claims rest on the proposed architecture and experimental outcomes rather than definitional equivalence or fitted-input renaming.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 3 invented entities

Review performed on abstract only; the paper introduces three new technical components whose independent validation cannot be checked from the given text.

axioms (1)
  • domain assumption LoRA-style adaptations can be applied to diffusion models for concept customization
    Standard background assumption in the custom diffusion model literature referenced by the abstract.
invented entities (3)
  • Attribute-decoupled LoRA (AD-LoRA) module no independent evidence
    purpose: Decouple concept-specific attributes to mitigate catastrophic forgetting
    New module introduced in the paper; no independent evidence supplied in abstract.
  • relevance-guided AD-LoRA aggregation strategy no independent evidence
    purpose: Leverage inter-task correlations while preserving per-task attributes
    New aggregation strategy proposed in the paper; no independent evidence supplied in abstract.
  • controllable regional context synthesis strategy no independent evidence
    purpose: Ensure semantic independence and smooth boundaries in multi-concept composition
    New synthesis strategy proposed in the paper; no independent evidence supplied in abstract.

pith-pipeline@v0.9.1-grok · 5759 in / 1454 out tokens · 45885 ms · 2026-06-28T06:41:24.657783+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

92 extracted references · 8 canonical work pages · 3 internal anchors

  1. [1]

    Il2m: Class incremental learning with dual memory,

    E. Belouadah and A. Popescu, “Il2m: Class incremental learning with dual memory,” inICCV, 2019, pp. 583–592

    2019

  2. [2]

    Stable Video Diffusion: Scaling Latent Video Diffusion Models to Large Datasets

    A. Blattmann, T. Dockhorn, S. Kulalet al., “Stable video diffusion: Scaling latent video diffusion models to large datasets,”arXiv preprint arXiv:2311.15127, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  3. [3]

    Align your latents: High-resolution video synthesis with latent diffusion models,

    A. Blattmann, R. Rombach, H. Ling, T. Dockhorn, S. W. Kim, S. Fidler, and K. Kreis, “Align your latents: High-resolution video synthesis with latent diffusion models,” inCVPR, June 2023, pp. 22 563–22 575

    2023

  4. [4]

    Attend- and-excite: Attention-based semantic guidance for text-to-image diffusion models,

    H. Chefer, Y. Alaluf, Y. Vinker, L. Wolf, and D. Cohen-Or, “Attend- and-excite: Attention-based semantic guidance for text-to-image diffusion models,”ACM Transactions on Graphics, vol. 42, no. 4, jul 2023

    2023

  5. [5]

    Disenstudio: Customized multi-subject text-to-video generation with disentangled spatial control,

    H. Chen, X. Wang, Y. Zhang, Y. Zhou, Z. Zhang, S. Tang, and W. Zhu, “Disenstudio: Customized multi-subject text-to-video generation with disentangled spatial control,” inACM MM, 2024

    2024

  6. [6]

    Any- door: Zero-shot object-level image customization,

    X. Chen, L. Huang, Y. Liu, Y. Shen, D. Zhao, and H. Zhao, “Any- door: Zero-shot object-level image customization,”arxiv preprint arxiv:2307.09481, 2023

    work page arXiv 2023

  7. [7]

    Dynasyn: Multi-subject personal- ization enabling dynamic action synthesis,

    Y. Choi, C. Park, and S. J. Baek, “Dynasyn: Multi-subject personal- ization enabling dynamic action synthesis,”AAAI, vol. 39, no. 3, pp. 2564–2572, Apr. 2025

    2025

  8. [8]

    Be your- self: Bounded attention for multi-subject text-to-image generation,

    O. Dahary, O. Patashnik, K. Aberman, and D. Cohen-Or, “Be your- self: Bounded attention for multi-subject text-to-image generation,” inECCV, 2024, pp. 432–448

    2024

  9. [9]

    No one left behind: Real-world federated class-incremental learning,

    J. Dong, H. Li, Y. Cong, G. Sun, Y. Zhang, and L. Van Gool, “No one left behind: Real-world federated class-incremental learning,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 46, no. 4, pp. 2054–2070, 2024

    2054

  10. [10]

    How to continually adapt text-to-image diffusion models for flexible customization?

    J. Dong, W. Liang, H. Li, D. Zhang, M. Cao, H. Ding, S. Khan, and F. S. Khan, “How to continually adapt text-to-image diffusion models for flexible customization?” inNeurIPS, vol. 37, 2024, pp. 130 057–130 083

    2024

  11. [11]

    Federated class-incremental learning,

    J. Dong, L. Wang, Z. Fang, G. Sun, S. Xu, X. Wang, and Q. Zhu, “Federated class-incremental learning,” inCVPR, June 2022, pp. 10 164–10 173

    2022

  12. [12]

    Dytox: Transformers for continual learning with dynamic token expansion,

    A. Douillard, A. Ramé, G. Couairon, and M. Cord, “Dytox: Transformers for continual learning with dynamic token expansion,” inCVPR, June 2022, pp. 9285–9295

    2022

  13. [13]

    Scaling rectified flow transformers for high-resolution image synthesis,

    P . Esser, S. Kulal, A. Blattmann, R. Entezari, J. Müller, H. Saini et al., “Scaling rectified flow transformers for high-resolution image synthesis,” inICML, 2024

    2024

  14. [14]

    An image is worth one word: Personalizing text-to-image generation using textual inversion,

    R. Gal, Y. Alaluf, Y. Atzmon, O. Patashnik, A. H. Bermano, G. Chechik, and D. Cohen-or, “An image is worth one word: Personalizing text-to-image generation using textual inversion,” inICLR, 2023

    2023

  15. [15]

    Phasemax: Convex phase retrieval via basis pursuit,

    T. Goldstein and C. Studer, “Phasemax: Convex phase retrieval via basis pursuit,”IEEE Transactions on Information Theory, vol. 64, no. 4, pp. 2675–2689, 2018

    2018

  16. [16]

    Vector quantized diffusion model for text-to-image synthesis,

    S. Gu, D. Chen, J. Bao, F. Wen, B. Zhang, D. Chen, L. Yuan, and B. Guo, “Vector quantized diffusion model for text-to-image synthesis,” inCVPR, June 2022, pp. 10 696–10 706

    2022

  17. [17]

    Mix- of-show: Decentralized low-rank adaptation for multi-concept customization of diffusion models,

    Y. Gu, X. Wang, J. Z. Wu, Y. Shi, C. Yunpeng, Z. Fanet al., “Mix- of-show: Decentralized low-rank adaptation for multi-concept customization of diffusion models,” inNeurIPS, 2023

    2023

  18. [18]

    Conceptguard: Continual personalized text- to-image generation with forgetting and confusion mitigation,

    Z. Guo and T. Jin, “Conceptguard: Continual personalized text- to-image generation with forgetting and confusion mitigation,” in CVPR, June 2025, pp. 2945–2954

    2025

  19. [19]

    Svdiff: Compact parameter space for diffusion fine-tuning,

    L. Han, Y. Li, H. Zhang, P . Milanfar, D. Metaxas, and F. Yang, “Svdiff: Compact parameter space for diffusion fine-tuning,” in ICCV, 2023, pp. 7289–7300

    2023

  20. [20]

    Cameractrl: Enabling camera control for video diffusion models,

    H. He, Y. Xu, Y. Guo, G. Wetzstein, B. Dai, H. Li, and C. Yang, “Cameractrl: Enabling camera control for video diffusion models,” inICLR, 2025

    2025

  21. [21]

    Streamingt2v: Consistent, dynamic, and extendable long video generation from text,

    R. Henschel, L. Khachatryan, H. Poghosyan, D. Hayrapetyanet al., “Streamingt2v: Consistent, dynamic, and extendable long video generation from text,” inCVPR, June 2025, pp. 2568–2577

    2025

  22. [22]

    Classifier-free diffusion guidance,

    J. Ho and T. Salimans, “Classifier-free diffusion guidance,” in NeurIPS 2021 Workshop on Deep Generative Models and Downstream Applications, 2021

    2021

  23. [23]

    LoRA: Low-rank adaptation of large language models,

    E. J. Hu, yelong shen, P . Wallis, Z. Allen-Zhu, Y. Li, S. Wang, L. Wang, and W. Chen, “LoRA: Low-rank adaptation of large language models,” inICLR, 2022

    2022

  24. [24]

    Turbo3d: Ultra-fast text-to-3d generation,

    H. Hu, T. Yin, F. Luan, Y. Hu, H. Tan, Z. Xu, S. Bi, S. Tulsiani, and K. Zhang, “Turbo3d: Ultra-fast text-to-3d generation,” inCVPR, June 2025, pp. 23 668–23 678

    2025

  25. [25]

    Storyagent: Customized storytelling video generation via multi- agent collaboration,

    P . Hu, J. Jiang, J. Chen, M. Han, S. Liao, X. Chang, and X. Liang, “Storyagent: Customized storytelling video generation via multi- agent collaboration,”arXiv preprint arXiv:2411.04925, 2024

    work page arXiv 2024

  26. [26]

    Videomage: Multi-subject and motion customization of text-to-video diffusion models,

    C.-P . Huang, Y.-S. Wu, H.-K. Chung, K.-P . Chang, F.-E. Yang, and Y.- C. F. Wang, “Videomage: Multi-subject and motion customization of text-to-video diffusion models,” inCVPR, June 2025, pp. 17 603– 17 612

    2025

  27. [27]

    Unicanvas: Affordance- aware unified real image editing via customized text-to-image generation,

    J. Jin, Y. Shen, X. Zhao, Z. Fu, and J. Yang, “Unicanvas: Affordance- aware unified real image editing via customized text-to-image generation,”International Journal of Computer Vision, vol. 133, pp. 3456–3480, 01 2025

    2025

  28. [28]

    Elucidating the design space of diffusion-based generative models,

    T. Karras, M. Aittala, T. Aila, and S. Laine, “Elucidating the design space of diffusion-based generative models,” inNeurIPS, 2022

    2022

  29. [29]

    Overcoming catastrophic forgetting in neural networks,

    J. Kirkpatrick, R. Pascanu, N. Rabinowitzet al., “Overcoming catastrophic forgetting in neural networks,”Proceedings of the National Academy of Sciences, vol. 114, no. 13, pp. 3521–3526, 2017

    2017

  30. [30]

    Multi-concept customization of text-to-image diffusion,

    N. Kumari, B. Zhang, R. Zhang, E. Shechtman, and J.-Y. Zhu, “Multi-concept customization of text-to-image diffusion,” inCVPR, 2023

    2023

  31. [31]

    B. F. Labs, “Flux,” https://github.com/black-forest-labs/flux, 2024

    2024

  32. [32]

    Blip: Bootstrapping language- image pre-training for unified vision-language understanding and generation,

    J. Li, D. Li, C. Xiong, and S. Hoi, “Blip: Bootstrapping language- image pre-training for unified vision-language understanding and generation,” inICML, 2022, pp. 12 888–12 900

    2022

  33. [33]

    Tuning-free image customization with image and text guidance,

    P . Li, Q. Nie, Y. Chen, X. Jiang, K. Wu, Y. Lin, Y. Liu, J. Peng, C. Wang, and F. Zheng, “Tuning-free image customization with image and text guidance,” inECCV, 2024, pp. 233–250

    2024

  34. [34]

    Motrans: Customized motion transfer with text-driven video diffusion models,

    X. Li, X. Jia, Q. Wang, H. Diao, mengmeng Ge, P . Li, Y. He, and H. Lu, “Motrans: Customized motion transfer with text-driven video diffusion models,” inACM MM, 2024

    2024

  35. [35]

    Gligen: Open-set grounded text-to-image generation,

    Y. Li, H. Liu, Q. Wu, F. Mu, J. Yang, J. Gao, C. Li, and Y. J. Lee, “Gligen: Open-set grounded text-to-image generation,” inCVPR, June 2023, pp. 22 511–22 521

    2023

  36. [36]

    Learning without forgetting,

    Z. Li and D. Hoiem, “Learning without forgetting,”IEEE Transac- tions on Pattern Analysis and Machine Intelligence, vol. 40, no. 12, pp. 2935–2947, 2017

    2017

  37. [37]

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

    C.-H. Lin, J. Gao, L. Tang, T. Takikawa, X. Zenget al., “Magic3d: High-resolution text-to-3d content creation,” inCVPR, June 2023, pp. 300–309

    2023

  38. [38]

    Mu- seummaker: Continual style customization without catastrophic forgetting,

    C. Liu, G. Sun, W. Liang, J. Dong, C. Qin, and Y. Cong, “Mu- seummaker: Continual style customization without catastrophic forgetting,”IEEE Transactions on Image Processing, vol. 34, pp. 2499– 2512, 2025

    2025

  39. [39]

    Make-your-3d: Fast and consistent subject-driven 3d content generation,

    F. Liu, H. Wang, W. Chen, H. Sun, and Y. Duan, “Make-your-3d: Fast and consistent subject-driven 3d content generation,” inECCV, 2024, pp. 389–406

    2024

  40. [40]

    Dora: weight-decomposed low-rank adaptation,

    S.-Y. Liu, C.-Y. Wang, H. Yin, P . Molchanov, Y.-C. F. Wang, K.-T. Cheng, and M.-H. Chen, “Dora: weight-decomposed low-rank adaptation,” inICML, 2024

    2024

  41. [41]

    C-CLIP: Multimodal continual learning for vision-language model,

    W. Liu, F. Zhu, L. Wei, and Q. Tian, “C-CLIP: Multimodal continual learning for vision-language model,” inICLR, 2025

    2025

  42. [42]

    Customizable image synthesis with multiple subjects,

    Z. Liu, Y. Zhang, Y. Shen, K. Zheng, K. Zhu, R. Feng, Y. Liuet al., “Customizable image synthesis with multiple subjects,” inNeurIPS, 2023. 17

    2023

  43. [43]

    Coarse-to-fine latent diffusion for pose-guided person image synthesis,

    Y. Lu, M. Zhang, A. J. Ma, X. Xie, and J. Lai, “Coarse-to-fine latent diffusion for pose-guided person image synthesis,” inCVPR, June 2024, pp. 6420–6429

    2024

  44. [44]

    Progressive rendering distillation: Adapting stable diffusion for instant text- to-mesh generation without 3d data,

    Z. Ma, X. Liang, R. Wu, X. Zhu, Z. Lei, and L. Zhang, “Progressive rendering distillation: Adapting stable diffusion for instant text- to-mesh generation without 3d data,” inCVPR, June 2025, pp. 11 036–11 050

    2025

  45. [45]

    Representational continuity for unsupervised continual learning,

    D. Madaan, J. Yoon, Y. Li, Y. Liu, and S. J. Hwang, “Representational continuity for unsupervised continual learning,” inICLR, 2022

    2022

  46. [46]

    Lt3sd: Latent trees for 3d scene diffusion,

    Q. Meng, L. Li, M. Nießner, and A. Dai, “Lt3sd: Latent trees for 3d scene diffusion,” inCVPR, June 2025, pp. 650–660

    2025

  47. [47]

    T2i-adapter: Learning adapters to dig out more controllable ability for text-to-image diffusion models,

    C. Mou, X. Wang, L. Xie, Y. Wu, J. Zhang, Z. Qi, Y. Shan, and X. Qie, “T2i-adapter: Learning adapters to dig out more controllable ability for text-to-image diffusion models,” inAAAI, 2024

    2024

  48. [48]

    Dream- matcher: Appearance matching self-attention for semantically- consistent text-to-image personalization,

    J. Nam, H. Kim, D. Lee, S. Jin, S. Kim, and S. Chang, “Dream- matcher: Appearance matching self-attention for semantically- consistent text-to-image personalization,” inCVPR, June 2024, pp. 8100–8110

    2024

  49. [49]

    Shapewords: Guiding text-to-image synthesis with 3d shape-aware prompts,

    D. Petrov, P . Goyal, D. Shivashok, Y. Tao, M. Averkiou, and E. Kalogerakis, “Shapewords: Guiding text-to-image synthesis with 3d shape-aware prompts,” inCVPR, June 2025, pp. 13 305–13 314

    2025

  50. [50]

    SDXL: Improving latent diffusion models for high-resolution image synthesis,

    D. Podell, Z. English, K. Lacey, A. Blattmann, T. Dockhorn, J. Müller, J. Penna, and R. Rombach, “SDXL: Improving latent diffusion models for high-resolution image synthesis,” inICLR, 2024

    2024

  51. [51]

    Dreamfusion: Text-to-3d using 2d diffusion,

    B. Poole, A. Jain, J. T. Barron, and B. Mildenhall, “Dreamfusion: Text-to-3d using 2d diffusion,” inICLR, 2023

    2023

  52. [52]

    Apply hierarchical-chain-of-generation to complex attributes text-to-3d generation,

    Y. Qin, Z. Xu, and Y. Liu, “Apply hierarchical-chain-of-generation to complex attributes text-to-3d generation,” inCVPR, June 2025, pp. 18 521–18 530

    2025

  53. [53]

    Dream- booth3d: Subject-driven text-to-3d generation,

    A. Raj, S. Kaza, B. Poole, M. Niemeyer, N. Ruizet al., “Dream- booth3d: Subject-driven text-to-3d generation,” inICCV, October 2023, pp. 2349–2359

    2023

  54. [54]

    Hierarchical Text-Conditional Image Generation with CLIP Latents

    A. Ramesh, P . Dhariwal, A. Nichol, C. Chu, and M. Chen, “Hier- archical text-conditional image generation with clip latents,”arxiv preprint arxiv:2204.06125, 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  55. [55]

    High-resolution image synthesis with latent diffusion models,

    R. Rombach, A. Blattmann, D. Lorenz, P . Esser, and B. Ommer, “High-resolution image synthesis with latent diffusion models,” in CVPR, 2022, pp. 10 684–10 695

    2022

  56. [56]

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

    N. Ruiz, Y. Li, V . Jampani, Y. Pritch, M. Rubinstein, and K. Aberman, “Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation,” inCVPR, 2023

    2023

  57. [57]

    Hyperdreambooth: Hypernet- works for fast personalization of text-to-image models,

    N. Ruiz, Y. Li, V . Jampani, W. Wei, T. Hou, Y. Pritch, N. Wadhwa, M. Rubinstein, and K. Aberman, “Hyperdreambooth: Hypernet- works for fast personalization of text-to-image models,” inCVPR, June 2024, pp. 6527–6536

    2024

  58. [58]

    Photorealistic text-to-image diffusion models with deep language understanding,

    C. Saharia, W. Chan, S. Saxena, L. Li, J. Whang, E. Dentonet al., “Photorealistic text-to-image diffusion models with deep language understanding,” inNeurIPS, 2022

    2022

  59. [59]

    Fast high-resolution image synthesis with latent adversarial diffusion distillation,

    A. Sauer, F. Boesel, T. Dockhorn, A. Blattmann, P . Esser, and R. Rom- bach, “Fast high-resolution image synthesis with latent adversarial diffusion distillation,” inSIGGRAPH Asia 2024 Conference Papers, 2024

    2024

  60. [60]

    Continual diffusion: Continual customization of text-to-image diffusion with c-lora,

    J. S. Smith, Y.-C. Hsu, L. Zhang, T. Hua, Z. Kira, Y. Shen, and H. Jin, “Continual diffusion: Continual customization of text-to-image diffusion with c-lora,”Transactions on Machine Learning Research, 2024

    2024

  61. [61]

    Multidreamer3d: Multi-concept 3d customization with concept-aware diffusion guidance,

    W. Song, S. Chang, and J. Yoo, “Multidreamer3d: Multi-concept 3d customization with concept-aware diffusion guidance,”arXiv preprint arXiv:2501.13449, 2025

    work page arXiv 2025

  62. [62]

    Create your world: Lifelong text-to-image diffusion,

    G. Sun, W. Liang, J. Dong, J. Li, Z. Ding, and Y. Cong, “Create your world: Lifelong text-to-image diffusion,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 46, no. 9, pp. 6454– 6470, 2024

    2024

  63. [63]

    Lgm: Large multi-view gaussian model for high-resolution 3d content creation,

    J. Tang, Z. Chen, X. Chenet al., “Lgm: Large multi-view gaussian model for high-resolution 3d content creation,” inECCV. Springer, 2024

    2024

  64. [64]

    Falcon: Fairness learning via contrastive attention approach to continual semantic scene understanding,

    T.-D. Truong, U. Prabhu, B. Raj, J. Cothren, and K. Luu, “Falcon: Fairness learning via contrastive attention approach to continual semantic scene understanding,” inCVPR, June 2025, pp. 15 065– 15 075

    2025

  65. [65]

    Anti-dreambooth: Protecting users from personalized text-to-image synthesis,

    T. Van Le, H. Phung, T. H. Nguyen, Q. Dao, N. N. Tran, and A. Tran, “Anti-dreambooth: Protecting users from personalized text-to-image synthesis,” inICCV, 2023, pp. 2116–2127

    2023

  66. [66]

    Dualreal: Adaptive joint training for lossless identity-motion fusion in video customization,

    W. Wang, M. Huang, Y. Tu, and Z. Mao, “Dualreal: Adaptive joint training for lossless identity-motion fusion in video customization,” inICCV, October 2025

    2025

  67. [67]

    MS-diffusion: Multi-subject zero-shot image personalization with layout guid- ance,

    X. Wang, S. Fu, Q. Huang, W. He, and H. Jiang, “MS-diffusion: Multi-subject zero-shot image personalization with layout guid- ance,” inICLR, 2025

    2025

  68. [68]

    Lavie: High-quality video generation with cascaded latent diffusion models,

    Y. Wang, X. Chen, X. Ma, S. Zhouet al., “Lavie: High-quality video generation with cascaded latent diffusion models,”International Journal of Computer Vision, 2025

    2025

  69. [69]

    Sigstyle: Signature style transfer via personalized text-to-image models,

    Y. Wang, T. Bai, X. Xie, Z. Yi, Y. Wang, and R. Ma, “Sigstyle: Signature style transfer via personalized text-to-image models,” AAAI, vol. 39, no. 8, pp. 8051–8059, Apr. 2025

    2025

  70. [70]

    Dual- prompt: Complementary prompting for rehearsal-free continual learning,

    Z. Wang, Z. Zhang, S. Ebrahimi, R. Sun, H. Zhanget al., “Dual- prompt: Complementary prompting for rehearsal-free continual learning,” inECCV, 2022, p. 631–648

    2022

  71. [71]

    Dream video: Composing your dream videos with customized subject and motion,

    Y. Wei, S. Zhang, Z. Qing, H. Yuan, Z. Liu, Y. Liu, Y. Zhang, J. Zhou, and H. Shan, “Dream video: Composing your dream videos with customized subject and motion,” inCVPR, 2024, pp. 6537–6549

    2024

  72. [72]

    Ouroboros3d: Image-to-3d generation via 3d-aware recursive diffusion,

    H. Wen, Z. Huang, Y. Wang, X. Chen, and L. Sheng, “Ouroboros3d: Image-to-3d generation via 3d-aware recursive diffusion,” inCVPR, 2025, pp. 21 631–21 641

    2025

  73. [73]

    Synthetic data is an elegant gift for continual vision-language models,

    B. Wu, W. Shi, J. Wang, and M. Ye, “Synthetic data is an elegant gift for continual vision-language models,” inCVPR, June 2025, pp. 2813–2823

    2025

  74. [74]

    Core: Context-regularized text embedding learning for text-to-image personalization,

    F. Wu, Y. Pang, J. Zhang, L. Pang, J. Yin, B. Zhao, Q. Li, and X. Mao, “Core: Context-regularized text embedding learning for text-to-image personalization,” inAAAI, 2025, pp. 8377–8385

    2025

  75. [75]

    Motionbooth: Motion-aware customized text-to-video generation,

    J. Wu, X. Li, Y. Zeng, J. Zhang, Q. Zhou, Y. Li, Y. Tong, and K. Chen, “Motionbooth: Motion-aware customized text-to-video generation,” inNeurIPS, 2024

    2024

  76. [76]

    Improved video vae for latent video diffusion model,

    P . Wu, K. Zhu, Y. Liu, L. Zhao, W. Zhai, Y. Cao, and Z.-J. Zha, “Improved video vae for latent video diffusion model,” inCVPR, June 2025, pp. 18 124–18 133

    2025

  77. [77]

    Customcrafter: Customized video genera- tion with preserving motion and concept composition abili- ties,

    T. Wu, Y. Zhang, X. Wang, X. Zhouet al., “Customcrafter: Cus- tomized video generation with preserving motion and concept composition abilities,”arXiv preprint arXiv:2408.13239, 2024

    work page arXiv 2024

  78. [78]

    Mixture of loRA experts,

    X. Wu, S. Huang, and F. Wei, “Mixture of loRA experts,” inICLR, 2024

    2024

  79. [79]

    Sana: Efficient high-resolution image synthesis with linear diffusion transformer,

    E. Xie, J. Chen, J. Chen, H. Cai, H. Tang, Y. Lin, Z. Zhanget al., “Sana: Efficient high-resolution image synthesis with linear diffusion transformer,” inICLR, 2024

    2024

  80. [80]

    Dreamvton: Customizing 3d virtual try-on with personalized diffusion models,

    Z. Xie, H. Dong, Y. Gao, Z. Ma, and X. Liang, “Dreamvton: Customizing 3d virtual try-on with personalized diffusion models,” inACM MM, 2024, p. 10784–10793

    2024

Showing first 80 references.