Pith. sign in

REVIEW 4 major objections 6 minor 42 references

PIDiff: Image Customization for Personalized Identities with Diffusion Models

T0 review · 4 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read PIDiff claims that feeding StyleGAN's W+ latent vectors through a dedicated cross-attention block preserves a person's identity in text-to-image generation better than text embeddings or CLIP image patches, while keeping the pretrained…

desk verdict PIDiff adds a sensible sequential cross-attention block to W+-based identity customization, but its quantitative claims are undermined by evaluation choices that favor the method. read the letter →

arxiv 2505.05081 v2 pith:J3VWJHU4 submitted 2025-05-08 cs.CV

Jinyu Gu , Haipeng Liu , Meng Wang , Yang Wang This is my paper · ORCID
classification cs.CV
keywords personalizedtext-to-imagegenerationidentitypreservationW+latentspacediffusionmodelfine-tuningcross-attentionstyleeditingGANinversionface
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

This paper proposes PIDiff, a fine-tuned diffusion model that represents a person's identity with a StyleGAN W+ latent vector rather than with a text embedding or a CLIP image patch. The paper argues that prior visual-prompt methods let identity and background entangle, so generated faces lose detail and the reference scene leaks into the output; PIDiff instead maps W+ codes through a Visual Guidance Module and injects them with a Style Cross-Attention block placed after the text cross-attention block. Only the new attention projection matrices and the mapping network are trained, for a few hundred steps, so the pretrained Stable Diffusion keeps its ability to follow text. On the authors' balanced 27-identity dataset, PIDiff reports the highest identity similarity among the compared methods (ID 0.3112) while matching the best text-image alignment (CLIP-T 0.1938), and it can edit style by concatenating coarse and fine W+ segments. The intended consequence is that personalized identity generation can change background, pose, clothing, and style according to the prompt without losing the person's face.

What carries the argument

The central mechanism is the Style Cross-Attention (SCA) block inserted after each text cross-attention block. It takes the text-conditioned hidden state as its query and the W+-derived visual embeddings as its keys and values, so the visual prompt binds to the region the text attention has already distinguished rather than spreading over the whole background. The Visual Guidance Module supplies those visual embeddings by splitting the 18x512 W+ code into four groups of latent codes, mapping each group through a separate layer into a 768-dimensional token, and concatenating the results, which preserves the coarse-to-fine structure of the W+ space. Only the SCA projection matrices and the mapping network are trainable, and a scalar lambda controls visual-prompt strength during inference, with the paper choosing lambda=0.4 as the balance point.

What would settle it

Take faces photographed against many different backgrounds, invert each cropped and aligned face to W+, and decode the W+ code with a StyleGAN generator: if the decoded image reconstructs background or scene structure rather than only the face, then scene information does leak into the visual prompt. Equivalently, measure the overlap, such as IoU, between PIDiff's visual-prompt attention map and the face segmentation mask across prompts; if that overlap drops sharply on cluttered backgrounds, the localization claim would be contradicted.

Watch

Extended reading notes

Core claim

The central claim is that identity information can be cleanly separated from background and scene information by representing a face as a W+ latent vector from StyleGAN rather than as text or CLIP image-patch embeddings. PIDiff crops and aligns the input face, inverts it with the e4e encoder to an 18x512 W+ code, and maps that code through a Visual Guidance Module into four 768-dimensional visual tokens. The tokens enter the diffusion U-Net through a Style Cross-Attention block placed after each text cross-attention block, with the text-conditioned hidden state serving as the query; this ordering is what confines the visual prompt to the face region. During training only the new SCA projection matrices and the mapping network are updated, with random W+ noise and a 0.5 drop probability, so the pretrained Stable Diffusion keeps its text-to-image abilities. On the paper's 27-identity dataset, PIDiff reports identity similarity 0.3112, LPIPS 0.5936, and CLIP-T 0.1938; W+ Adapter, the closest rival, reports 0.2668, 0.6774, and 0.1935, and the other compared baselines score lower on identity.

Load-bearing premise

The whole improvement depends on the claim that the e4e encoder's W+ code for a cropped, aligned face captures identity without also carrying background or scene content; the paper supports this with attention-map pictures, not with a quantitative test of disentanglement.

Editorial extensions

If this is right

  • Using W+ prompts instead of text or CLIP image-patch prompts raises identity similarity from 0.2858 (IP-Adapter) and 0.2668 (W+ Adapter) to 0.3112 on the paper's dataset, with CLIP-T essentially tied at 0.1938.
  • Because only the SCA projection matrices and the mapping network are trainable, PIDiff preserves the pretrained model's text-following ability, so per-identity tuning stays light and does not destroy the base model.
  • W+ vectors are organized coarse-to-fine, so combining layers 1-9 of one identity with layers 10-18 of another edits visible style while keeping identity, and adding a style image to the training set gives a similar effect.
  • Six training images gave the best results; four overfit and eight or ten make identity harder to learn.
  • A lambda of 0.4 in SCA balances identity and text alignment; smaller lambda loses identity, while larger lambda drops CLIP-T sharply.

Reading between the lines

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

  • Inference: The paper's advantage would be more convincing if W+ disentanglement were measured directly; decoding the inverted W+ code and checking whether background structure is recoverable would settle whether identity and scene information are really separated.
  • Inference: SCA's query-from-text-cross-attention design is a general recipe: any condition that should bind to a region, not just a face, could be injected the same way, potentially extending PIDiff to objects, logos, or scene elements.
  • Inference: The global lambda scalar could be made layer-specific or timestep-dependent; the reported lambda=0.4 result suggests a schedule that raises identity weight at early denoising steps and lowers it later might improve both metrics beyond the single trade-off point.
  • Inference: The 27-identity dataset is small and self-constructed; numbers like ID 0.3112 are meaningful relative to the baselines in the table, but the general ranking would be on firmer ground if reproduced on an independent multi-image-per-identity benchmark.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes PIDiff, a fine-tuning-based method for personalized identity text-to-image generation. The method maps a cropped and aligned face image to the StyleGAN W+ space using the e4e encoder, transforms the resulting w+ vector into visual tokens through a mapping network, and injects these tokens into Stable Diffusion through a newly designed Style Cross-Attention (SCA) block placed after each text cross-attention block. Only the SCA projection matrices and the mapping network are trainable during a short per-identity fine-tuning stage. The paper also introduces a small dataset of 27 identities with 10 images each and reports that PIDiff achieves the highest identity similarity (ID 0.3112) and the best or tied text alignment (CLIP-T 0.1938) on this dataset, along with qualitative results and an ablation study that selects the number of training images and the SCA weight λ. The central claim is that PIDiff outperforms state-of-the-art methods while better preserving identity and maintaining text-image consistency.

Significance. If the central claim holds, the paper makes a useful contribution by demonstrating that a StyleGAN W+ visual prompt can be integrated into a diffusion model for identity-preserving generation, and that a serial cross-attention structure after text cross-attention can improve visual-prompt localization. The idea of limiting trainable parameters to the SCA projections and mapping network is sensible and aligns with recent parameter-efficient fine-tuning practice. The paper also makes a positive effort to construct a dataset with balanced demographic categories. The strengths are the architectural proposal, the attention-map analysis, and the style-editing demonstration via coarse-to-fine w+ vector concatenation. However, the quantitative evidence as currently presented is not sufficient to establish the claimed superiority over prior methods, for the reasons detailed in the major comments.

major comments (4)
  1. [Sec. 3.3.1, Table 1] The ID evaluation protocol is not comparable across the rows of Table 1. For methods that do not take a reference image at inference time (Textual Inversion, Custom Diffusion, DreamBooth, and possibly BLIP-Diffusion and VICO), the paper states that it selects the most similar facial image from the training images to compute the evaluation metrics, whereas for IP-Adapter, W+ Adapter, and PIDiff the ID is computed against a fixed reference image. This is a per-sample maximum over a 10-image gallery for some methods and a fixed-reference protocol for others, which can inflate ID scores for the reference-free methods and makes the cross-method comparison misleading. The authors should use the same fixed-reference scoring rule for all methods, or otherwise justify why the per-sample max is appropriate and show that the conclusions do not change under a uniform protocol.
  2. [Sec. 3.4.4 and Table 2] The reported quantitative superiority is partly circular because the two key hyperparameters are selected on the same benchmark that is then used for the final comparison. Section 3.4.4 states that λ=0.4 is chosen because it maximizes text-image semantic consistency and identity preservation, presumably on the same 12 prompts and 27 identities used in Table 1, and Section 3.4.2 selects six training images from the same Table 2 results. No held-out split or nested selection procedure is described. The authors should evaluate on a disjoint held-out set of identities and prompts, or at least report the selection procedure and show that the final ranking is stable under cross-validation.
  3. [Table 1] Table 1 reports no error bars, confidence intervals, or multiple-seed results. Several of the decisive margins are very small: the CLIP-T difference between PIDiff (0.1938) and W+ Adapter (0.1935) is 0.0003, and the LPIPS difference between PIDiff (0.5936) and IP-Adapter (0.5947) is 0.0011. Given the stochasticity of diffusion sampling and the small evaluation set, these differences are within plausible run-to-run noise. The authors should run each method and their own method with multiple random seeds and report mean ± std, and should state whether the reported numbers are averages or single runs.
  4. [Sec. 2.2.2, Fig. 3] The central mechanism claim that cropping and e4e inversion into W+ spaces disentangles identity information from background and other scene content is not quantitatively verified. The paper only provides attention-map visualizations in Fig. 3, which are suggestive but not a quantitative test. This assumption is load-bearing for the claimed advantage over W+ Adapter, because if scene information survives in W+ the visual prompt could still interfere with backgrounds and the SCA localization would fail. A concrete test would be, for example, measuring whether the generated background changes when the same identity is presented in different backgrounds, or comparing attention maps and output diversity when W+ coefficients that are not identity-specific are manipulated. This test should be added before the mechanism-based explanation of the advantage is accepted.
minor comments (6)
  1. [Fig. 3 and references] The figure labels do not match the bibliography: Fig. 3 cites "W+ adapter [12]" and "IP-Adapter [32]", but the corresponding references in the bibliography are [13] and [37]. All in-text citation numbers should be checked for consistency.
  2. [Sec. 2.3 and Sec. 3.4.2] The dataset description says each identity is represented by ten images, but the training-configuration analysis in Sec. 3.4.2 uses four, six, eight, or ten training images. It should be clarified how the remaining images are used, whether the same six images are selected for all identities, and whether the test prompts overlap with any training images.
  3. [References] References [20] and [21] appear to be the same paper (StylRes), and references [24] and [25] are the same CLIP paper with inconsistent venue capitalization ("PmLR" in [25]). These should be merged or corrected.
  4. [Notation] The paper uses both `𝑤+` and `w+` for the W+ latent space, sometimes within the same section (e.g., Sec. 2.2.2 and Fig. 1). The notation should be standardized throughout.
  5. [Sec. 3.4.2] The explanation that four training images cause the model to "overfit images and text prompts" while six images avoid overfitting is not self-evident and is not supported by the qualitative figure alone. Please clarify the expected behavior and, if possible, show quantitative evidence for the stated overfitting effect.
  6. [Sec. 2.3 and Sec. 3.3.1] The paper refers to providing the dataset and code in the supplementary material but does not state whether they will be publicly released or under what license. This information should be explicit for reproducibility.

Circularity Check

1 steps flagged · score 4.0 of 10

Headline superiority is partly tuned: λ=0.4 and the six-image training configuration are selected on the same benchmark that is later re-reported as the final comparison row.

  1. fitted input called prediction [Sec. 3.4.2, Sec. 3.4.4, Table 1, Table 2]
    "We choose to use six images for training. ... When the λ is larger than 0.4, CLIP-T rapidly decreases, while ID does not change significantly. Therefore, through experimental analysis, we choose 0.4 as an appropriate choice."

    Table 2 reports ID=0.3112 and CLIP-T=0.1938 for six training images with SCA, and Sec. 3.4.4 selects λ=0.4 by inspecting the ID/CLIP-T trade-off. Table 1 then gives the same values (ID 0.3112, LPIPS 0.5936, CLIP-T 0.1938) as the 'Ours' row on the same 12 prompts × 27 identities, with no held-out split described. The headline comparison is therefore a re-reporting of the configuration chosen to optimize those very metrics, not an independent test of a pre-specified method. The SCA-versus-PCA ablation provides some independent support for the architecture, so the circularity is partial rather than total.

full rationale

The core technical derivation is not circular: SCA is defined from text cross-attention outputs and W+ embeddings (Eqs. 3-5), Stable Diffusion and e4e are external components, and the SCA-vs-PCA ablation gives an independent, if small, test of the architectural choice. No load-bearing self-citation chain is present; the authors' own prior works are cited only as generic diffusion/optimization references, not as justifications of the identity-preservation claim. The e4e-disentanglement assumption is a mechanism risk but not a circular step, since W+ is not defined in terms of the downstream metric. The main circularity is at the evaluation level: the number of training images and λ are chosen by inspecting ID and CLIP-T on the same 12-prompt × 27-identity set later used in Table 1, and the final 'Ours' row simply re-reports the chosen configuration. The separate reference-scoring rule for reference-free baselines in Sec. 3.3.1 is a fairness concern, not a circularity, because it does not make PIDiff's output equal to its input. Overall this is partial circularity: the architecture choice has independent grounding, but the reported numerical superiority is partly forced by tuning on the evaluation set.

Assumptions & free parameters 3 free parameters · 4 assumptions · 1 invented entities

PIDiff rests on empirical claims about W+ disentanglement, partial fine-tuning preserving the base model, and the validity of the small benchmark. The main hyperparameters used in the final comparison are selected on the evaluation metrics themselves, which means part of the reported gain is a fitted result rather than an independent prediction.

free parameters (3)
  • SCA weight lambda = 0.4
    Set in Sec. 3.4.4 based on the ID and CLIP-T tradeoff on the evaluation set, then used for all reported main results.
  • number of training images per identity = 6
    Selected in Sec. 3.4.2 because it gives the best ID and CLIP-T values in Table 2; other counts overfit or underlearn identity.
  • probability of dropping visual and text embeddings during training = 0.5
    Chosen in Sec. 3.1 instead of the 0.05 used by IP-Adapter and W+ Adapter; the paper says more analysis is in the supplementary material.
assumptions (4)
  • domain assumption The e4e W+ representation of an aligned face crop is disentangled enough that background and scene content do not leak into the visual prompt.
    Sec. 2.2.2 relies on cropping and alignment to remove background interference before inversion; if this fails, the visual prompt still carries scene information.
  • domain assumption Fine-tuning only the SCA projection matrices and mapping network for 600 steps preserves the pretrained model's ability to follow arbitrary text prompts.
    Sec. 2.2.4 states this as a design principle but provides no controlled measurement of forgetting.
  • domain assumption ArcFace cosine similarity on MTCNN-aligned faces measures identity preservation fairly across different poses and backgrounds.
    Used as the primary identity metric in Sec. 3.2 with no calibration for generated images.
  • domain assumption A test set of 27 identities and 12 prompts per identity is sufficient to rank methods.
    Sec. 3.3.1 provides no error bars or significance tests, so the ranking could change with a larger or different benchmark.
invented entities (1)
  • Style Cross-Attention (SCA) block
    purpose: Injects visual prompts derived from W+ vectors into the denoising U-Net after the text cross-attention block.
    The only evidence is the internal ablation versus parallel cross-attention in Table 2; no external benchmark, theory, or released model confirms its advantage.

how reviews work

0 comments
Cite this review

Pith. "Pith review of PIDiff: Image Customization for Personalized Identities with Diffusion Models." pith.science (2026). https://pith.science/paper/J3VWJHU4

@misc{pith2026250505081,
  author       = {Pith},
  title        = {Pith review of: PIDiff: Image Customization for Personalized Identities with Diffusion Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/J3VWJHU4}},
  note         = {Machine review of arXiv:2505.05081}
}
read the original abstract

Text-to-image generation for personalized identities aims at incorporating the specific identity into images using a text prompt and an identity image. Based on the powerful generative capabilities of DDPMs, many previous works adopt additional prompts, such as text embeddings and CLIP image embeddings, to represent the identity information, while they fail to disentangle the identity information and background information. As a result, the generated images not only lose key identity characteristics but also suffer from significantly reduced diversity. To address this issue, previous works have combined the W+ space from StyleGAN with diffusion models, leveraging this space to provide a more accurate and comprehensive representation of identity features through multi-level feature extraction. However, the entanglement of identity and background information in in-the-wild images during training prevents accurate identity localization, resulting in severe semantic interference between identity and background. In this paper, we propose a novel fine-tuning-based diffusion model for personalized identities text-to-image generation, named PIDiff, which leverages the W+ space and an identity-tailored fine-tuning strategy to avoid semantic entanglement and achieves accurate feature extraction and localization. Style editing can also be achieved by PIDiff through preserving the characteristics of identity features in the W+ space, which vary from coarse to fine. Through the combination of the proposed cross-attention block and parameter optimization strategy, PIDiff preserves the identity information and maintains the generation capability for in-the-wild images of the pre-trained model during inference. Our experimental results validate the effectiveness of our method in this task.

Figures

Figures reproduced from arXiv: 2505.05081 by the authors.

Figure 1
Figure 1. Given face images with coarse-grained features (e.g., hair, skin color, face shape) and fine-grained features (facial [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of the proposed PIDiff. PIDiff consists of two modules: Visual Guidance Module(VGM) and the diffusion [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Comparison of Image Generation Results and Atten [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Illustration of Style Cross-Attention(SCA). SCA [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 6
Figure 6. Figure 6: Visualization of our proposed personalized identity [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: Qualitative Comparison between previous methods and PIDiff. PIDiff not only maintains identity features and [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: Comparative Analysis of Models Across Various [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]
Figure 9
Figure 9. Figure 9: Examples generated by the model [37] using CLIP-I as the image encoder. The facial features are overly fixed, [PITH_FULL_IMAGE:figures/full_fig_p007_9.png]
Figure 10
Figure 10. Figure 10: Visual comparisons of images generated by using different [PITH_FULL_IMAGE:figures/full_fig_p008_10.png]
Figure 11
Figure 11. Figure 11: Qualitative results of using different numbers of [PITH_FULL_IMAGE:figures/full_fig_p008_11.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 8 canonical work pages

  1. [1]

    Yogesh Balaji, Seungjun Nah, Xun Huang, Arash Vahdat, Jiaming Song, Qinsheng Zhang, Karsten Kreis, Miika Aittala, Timo Aila, Samuli Laine, et al. 2022. ediff-i: Text-to-image diffusion models with an ensemble of expert denoisers. arXiv preprint arXiv:2211.01324 (2022)

    arXiv 2022

  2. [2]

    Ahmet Canberk Baykal, Abdul Basit Anees, Duygu Ceylan, Erkut Erdem, Aykut Erdem, and Deniz Yuret. 2023. CLIP-guided StyleGAN Inversion for Text-driven Real Image Editing. ACM Transactions on Graphics 42, 5 (2023), 1–18

    work page 2023

  3. [3]

    Denis Bobkov, Vadim Titov, Aibek Alanov, and Dmitry Vetrov. 2024. The devil is in the details: Stylefeatureeditor for detail-rich stylegan inversion and high quality image editing. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 9337–9346

    work page 2024

  4. [4]

    Jiankang Deng, Jia Guo, Niannan Xue, and Stefanos Zafeiriou. 2019. Arcface: Additive angular margin loss for deep face recognition. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition . 4690–4699

    2019

  5. [5]

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

    arXiv 2022

  6. [6]

    Shaozhe Hao, Kai Han, Shihao Zhao, and Kwan-Yee K Wong. 2023. ViCo: Plug- and-play Visual Condition for Personalized Text-to-image Generation. arXiv preprint arXiv:2306.00971 (2023)

    arXiv 2023

  7. [7]

    Geoffrey Hinton, Oriol Vinyals, and Jeff Dean. 2015. Distilling the knowledge in a neural network. arXiv preprint arXiv:1503.02531 (2015)

    arXiv 2015

  8. [8]

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. 2020. Denoising diffusion probabilistic models. Advances in neural information processing systems 33 (2020), 6840–6851

    2020

Show all 42 references
  1. [9]

    Tero Karras. 2019. A Style-Based Generator Architecture for Generative Adver- sarial Networks. arXiv preprint arXiv:1812.04948 (2019)

    2019 arXiv

  2. [10]

    Nupur Kumari, Bingliang Zhang, Richard Zhang, Eli Shechtman, and Jun-Yan Zhu

  3. [11]

    Dongxu Li, Junnan Li, and Steven Hoi. 2024. Blip-diffusion: Pre-trained subject representation for controllable text-to-image generation and editing. Advances in Neural Information Processing Systems 36 (2024)

    2024

  4. [12]

    Hao Li, Mengqi Huang, Lei Zhang, Bo Hu, Yi Liu, and Zhendong Mao. 2024. Grad- ual residuals alignment: a dual-stream framework for GAN inversion and image attribute editing. In Proceedings of the AAAI Conference on Artificial Intelligence , Vol. 38. 3064–3072

    2024

  5. [13]

    Xiaoming Li, Xinyu Hou, and Chen Change Loy. 2024. When stylegan meets stable diffusion: a w+ adapter for personalized image generation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 2187–2196

    2024

  6. [14]

    Hongyu Liu, Yibing Song, and Qifeng Chen. 2023. Delving stylegan inversion for image editing: A foundation latent space viewpoint. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition . 10072–10082

    2023

  7. [15]

    Haipeng Liu, Yang Wang, Biao Qian, Meng Wang, and Yong Rui. 2024. Struc- ture matters: Tackling the semantic discrepancy in diffusion models for image inpainting. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 8038–8047

    2024

  8. [16]

    Haipeng Liu, Yang Wang, Meng Wang, and Yong Rui. 2022. Delving globally into texture and structure for image inpainting. In Proceedings of the 30th ACM International Conference on Multimedia . 1270–1278

    2022

  9. [17]

    Jing Long, Guanhua Ye, Tong Chen, Yang Wang, Meng Wang, and Hongzhi Yin. 2024. Diffusion-based cloud-edge-device collaborative learning for next POI recommendations. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining . 2026–2036

    2024

  10. [18]

    I Loshchilov. 2017. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101 (2017)

    2017 arXiv

  11. [19]

    Alex Nichol, Prafulla Dhariwal, Aditya Ramesh, Pranav Shyam, Pamela Mishkin, Bob McGrew, Ilya Sutskever, and Mark Chen. 2021. Glide: Towards photorealistic image generation and editing with text-guided diffusion models. arXiv preprint arXiv:2112.10741 (2021)

    2021 arXiv

  12. [21]

    Hamza Pehlivan, Yusuf Dalva, and Aysegul Dundar. 2023. Styleres: Transforming the residuals for real image editing with stylegan. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition . 1828–1837

    2023

  13. [22]

    Biao Qian, Yang Wang, Richang Hong, and Meng Wang. 2023. Adaptive data-free quantization. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 7960–7968

    2023

  14. [23]

    Biao Qian, Yang Wang, Richang Hong, and Meng Wang. 2023. Rethinking data- free quantization as a zero-sum game. In Proceedings of the AAAI conference on artificial intelligence, Vol. 37. 9489–9497

    2023

  15. [24]

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. 2021. Learning transferable visual models from natural language supervision. In International conference on machine learni...

    2021

  16. [25]

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. 2021. Learning transferable visual models from natural language supervision. In International conference on machine learni...

    2021

  17. [26]

    Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen

  18. [27]

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

    2022

  19. [28]

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. 2015. U-net: Convolu- tional networks for biomedical image segmentation. In Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceeding...

    2015

  20. [29]

    Nataniel Ruiz, Yuanzhen Li, Varun Jampani, Yael Pritch, Michael Rubinstein, and Kfir Aberman. 2023. Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition . 22500–22510

    2023

  21. [30]

    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. 2022. Photorealistic text-to-image diffusion models with deep language understanding. Advances in neural infor...

    2022

  22. [31]

    Jiaming Song, Chenlin Meng, and Stefano Ermon. 2020. Denoising diffusion implicit models. arXiv preprint arXiv:2010.02502 (2020)

    2020 arXiv

  23. [32]

    Omer Tov, Yuval Alaluf, Yotam Nitzan, Or Patashnik, and Daniel Cohen-Or. 2021. Designing an encoder for stylegan image manipulation. ACM Transactions on Graphics (TOG) 40, 4 (2021), 1–14

    2021

  24. [33]

    Tengfei Wang, Yong Zhang, Yanbo Fan, Jue Wang, and Qifeng Chen. 2022. High- fidelity gan inversion for image attribute editing. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition . 11379–11388

    2022

  25. [34]

    Yang Wang, Biao Qian, Haipeng Liu, Yong Rui, and Meng Wang. 2024. Unpacking the gap box against data-free knowledge distillation. IEEE Transactions on Pattern Analysis and Machine Intelligence (2024)

    2024

  26. [35]

    Guangxuan Xiao, Tianwei Yin, William T Freeman, Frédo Durand, and Song Han

  27. [36]

    Zeyue Xue, Guanglu Song, Qiushan Guo, Boxiao Liu, Zhuofan Zong, Yu Liu, and Ping Luo. 2024. Raphael: Text-to-image generation via large mixture of diffusion paths. Advances in Neural Information Processing Systems 36 (2024)

    2024

  28. [37]

    Hu Ye, Jun Zhang, Sibo Liu, Xiao Han, and Wei Yang. 2023. Ip-adapter: Text compatible image prompt adapter for text-to-image diffusion models. arXiv preprint arXiv:2308.06721 (2023)

    2023 arXiv

  29. [38]

    Kaipeng Zhang, Zhanpeng Zhang, Zhifeng Li, and Yu Qiao. 2016. Joint face detection and alignment using multitask cascaded convolutional networks. IEEE signal processing letters 23, 10 (2016), 1499–1503

    2016

  30. [39]

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang

  31. [2018]

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

    The unreasonable effectiveness of deep features as a perceptual metric. In Proceedings of the IEEE conference on computer vision and pattern recognition . 586–595

  32. [2022]

    arXiv preprint arXiv:2204.06125 1, 2 (2022), 3

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

    2022 arXiv

  33. [2023]

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

    Multi-concept customization of text-to-image diffusion. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition . 1931–1941

    1931

  34. [2024]

    International Journal of Computer Vision (2024), 1–20

    Fastcomposer: Tuning-free multi-subject image generation with localized attention. International Journal of Computer Vision (2024), 1–20

    2024

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.