Pith. sign in

REVIEW 3 major objections 6 minor 1 cited by

Domain Generalizable Portrait Style Transfer

T0 review · 3 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A portrait style transfer framework trained only on real photographs (about 28K images) can robustly transfer style between portraits from very different domains — photo, cartoon, sketch, animation, painting — while preserving identity…

desk verdict A practical, fast portrait style transfer system with a clever architecture, but the domain-generalization claim is under-supported and the Gram metric is mistargeted. read the letter →

arxiv 2507.04243 v2 pith:7VKTOJ37 submitted 2025-07-06 cs.CV cs.AI

Xinbo Wang , Wenju Xu , Qing Zhang , Wei-Shi Zheng This is my paper
classification cs.CVcs.AI
keywords portraitstyletransferdomaingeneralizationdiffusionmodelssemanticcorrespondenceAdaIN-WavelettransformControlNetadapterimage-to-imagetranslation
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 claims that portrait style transfer can be made domain-generalizable: a model trained only on real portrait photographs learns to transfer style between portraits from very different domains, including photos, cartoons, sketches, animations, and paintings. The method works in three steps: it computes dense semantic correspondence between the input and reference using features from a frozen diffusion model plus a small trainable semantic adapter; it warps the reference to align with the input; and it generates the output with a dual-conditional diffusion process that balances high-frequency structure from the input with stylized color and texture from the warped reference. The authors report that on both a real-photo benchmark and a mixed multi-domain dataset, the approach improves content preservation and identity retention while achieving style transfer quality at least comparable to prior art, using only about 28,000 training photos.

What carries the argument

The central machinery is the three-part synthesis of dense semantic correspondence and dual conditional diffusion. The correspondence part is a trainable semantic adapter attached to a frozen Stable Diffusion U-Net and CLIP image encoder; it produces feature maps whose correlation matrix (with channel-wise mean subtraction, equation 3) is used to warp the reference (equation 4). Two losses refine it: mask-warping loss (equation 5) and cyclic warping consistency loss (equation 6). The generation part is a dual-conditional diffusion model: a ControlNet that takes high-frequency components of the input obtained by Haar discrete wavelet transform as structure guidance (equations 7–8), and a style adapter that injects CLIP text and image features via decoupled cross-attention (equation 9). The initialization is an AdaIN-Wavelet transform that first applies AdaIN between the DDIM-inverted warped-reference latent and the input latent, then recombines the low-frequency band of the warped-reference latent with high-frequency bands of the AdaIN-blended latent via inverse wavelet transform (equations 10–11); a scalar gamma controls stylization strength by interpolating between the input and stylized latents (equation 12).

What would settle it

Measure the alignment accuracy of the warped reference against the input on a held-out cross-domain dataset where ground-truth matches or semantic masks exist (for example, face landmarks or segmentation); if the warped reference's mask IoU or landmark distance in sketch-to-photo or cartoon-to-photo pairs is significantly worse than in photo-to-photo pairs, the domain-generalization claim fails. A simpler check: replace the semantic adapter with frozen diffusion features only (the ablation of Section 5.3) and see whether cross-domain warping visibly misaligns eyes and mouths.

Watch

Extended reading notes

Core claim

The paper's central claim is that a single framework, trained on about 28,000 real portrait photos, can perform high-quality semantically aligned style transfer between portraits from arbitrary domains — photo, cartoon, sketch, animation, painting — without requiring paired training data from those domains. The key discovery is that dense semantic correspondence, extracted from a pre-trained Stable Diffusion U-Net and refined by a lightweight semantic adapter with mask-warping and cyclic-warping consistency losses, provides reliable cross-domain alignment. On top of this alignment, an AdaIN-Wavelet transform blends the low-frequency color and tone of the warped reference latent with the high-frequency detail of the input latent, giving an initial latent that balances stylization and content preservation. A dual-conditional diffusion model, combining a ControlNet fed with high-frequency wavelet coefficients (structure guidance) and a style adapter with decoupled cross-attention (style guidance), produces the final image. The paper reports that this design outperforms existing portrait style transfer methods on Gram loss, LPIPS, and identity distance metrics, especially on a mixed multi-domain dataset.

Load-bearing premise

The load-bearing premise is that the dense correspondence estimator, trained only on real photos and their segmentation masks, keeps matching facial parts accurately when the input and reference come from highly non-photorealistic domains like sketch, cartoon, or painting; if the warped reference is misaligned there, the rest of the pipeline inherits the error.

Editorial extensions

If this is right

  • Trained on real photos alone, the model can colorize grayscale and sketch portraits, modernize old photographs, and restyle real photos in sketch style, as shown in Figures 9–11.
  • The gamma parameter gives continuous control over stylization strength, and region-specific masks enable localized style transfer (hair, face, lips) without retraining.
  • The method is fast enough for practical use: about 6.97 seconds per 512x512 image on an RTX 4090, faster than the compared diffusion baselines.
  • The design suggests that domain-generalizable style transfer can be achieved without large-scale multi-domain training data, since the correspondence, structure, and style modules are trained or frozen on real photos.

Reading between the lines

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

  • A testable extension: adding a small set of synthetic or stylized portraits (sketches, cartoons) to the semantic-adapter training should improve cross-domain alignment accuracy without harming photo-domain performance; a reader could verify this with the released code.
  • The AdaIN-Wavelet latent blending is a general mechanism, not portrait-specific; it could transfer to other exemplar-based image synthesis tasks (relighting, makeup transfer, object recoloring) where a warped exemplar and content preservation trade off.
  • The method's dependence on Stable Diffusion and CLIP features implies that upgrading to newer foundation models may further improve robustness, but also inherits their biases, for instance regarding faces of certain demographics.
  • Because the paper's quantitative evaluation uses LPIPS and identity distance as proxies, a human-perception study with domain-specific raters would be a stronger test of whether the style transfer is truly faithful across domains.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper proposes a portrait style transfer framework, DGPST, that aims to generalize across image domains (photo, cartoon, sketch, animation, painting) while training only on roughly 28K real portrait photos from CelebAMask-HQ. The method first establishes dense semantic correspondence between an input portrait and a reference portrait using features from a frozen Stable Diffusion U-Net refined by a trainable semantic adapter, with mask-warping and cyclic-warping consistency losses. The reference is warped according to this correspondence. A dual-conditional diffusion model then generates the output: a ControlNet receives high-frequency wavelet components of the input for structure guidance, and a style adapter receives the warped reference for style guidance. The denoising trajectory starts from an AdaIN-Wavelet-initialized latent that combines low-frequency information from the warped reference latent with high-frequency information from the input latent. Experiments compare against several style-transfer baselines using Gram loss, LPIPS, and identity distance, and include qualitative results on cross-domain inputs. The paper claims state-of-the-art performance and domain generalizability, with code and models released.

Significance. If the empirical claims are substantiated, the paper makes a useful contribution: it demonstrates a way to achieve semantic-aware portrait style transfer across diverse artistic domains without training on those domains, by leveraging the representational power of a pre-trained diffusion model plus a lightweight adapter. The architecture is modular and the release of code and trained models supports reproducibility. The central idea of using diffusion features for dense semantic correspondence and then conditioning a diffusion generator on both structure and style is interesting and potentially impactful for portrait editing. However, the current quantitative evaluation does not support the strength of the stated claims because one of the headline metrics is defined in a way that does not evaluate the generated output, and the cross-domain generalization claim lacks direct correspondence-quality evidence. The contribution is therefore conditionally significant, pending fixes to the evaluation protocol and additional experiments.

major comments (3)
  1. [Section 5.1, Tables 1 and 2] The Gram loss is defined as the loss 'between the input image and the warped reference' (Section 5.1), not between the output and the reference or between the output and the input. As written, this metric does not measure the stylization of the generated image at all; a degenerate method that simply outputs the input would achieve zero Gram loss. The low Gram-loss values for the proposed method in Tables 1 and 2 thus do not support the claim of state-of-the-art stylization quality. Please compute a style metric that involves the output (e.g., Gram loss between output and reference) and report those numbers, or explain why the current definition is meaningful for the task.
  2. [Section 4.4, Eq. (15)] The second-stage training loss uses `c_cnt and c_sty both extracted from the input image`, meaning the dual-conditional diffusion model is trained only with the input's own style as the style condition. At test time, however, `c_sty` is extracted from the warped reference, which is a different image. This train/test mismatch is not analyzed or ablated. If the model learns to rely on content-specific features that are present only when the style prompt coincides with the input, its ability to transfer arbitrary reference styles would be limited. The paper should provide an ablation that trains the style adapter with pairs of distinct images (or at least with a style-augmented version of the input) and compare the resulting output quality.
  3. [Section 4.1, cross-domain generalization] The semantic adapter is trained only on CelebAMask-HQ real-photo pairs with real-photo masks and cyclic-warping losses, yet the paper's central claim is that the method 'robustly process portraits from various different domains.' The evidence for cross-domain correspondence accuracy is limited to a single qualitative correspondence visualization (Figure 4) and indirect downstream metrics (Table 2 on a mixed dataset). Because the warped reference is the foundation of both the AdaIN-Wavelet initialization and the style adapter, an inaccurate correspondence would degrade the whole pipeline exactly in the cross-domain regime. Please provide a quantitative evaluation of correspondence accuracy on cross-domain pairs (e.g., PCK or a similar metric) or a systematic qualitative study with multiple examples per domain, including failure cases.
minor comments (6)
  1. [Abstract and Section 5.1] The abstract states training on '30K portrait photos,' while Section 5.1 says the model is trained on the 28K training images of CelebAMask-HQ; please reconcile these numbers.
  2. [Eq. (3)] The notation `mu_{F_c0} in R^{HW}` is described as a mean vector 'along the channel dimension'; if the intent is a per-pixel mean across channels, the description should say so explicitly, as the current wording is ambiguous.
  3. [Section 4.3, Eq. (11)] It is not clear whether the wavelet transform is applied separately to `z_s_w` and `z_cs'` before taking their low- and high-frequency components; please specify the decomposition inputs.
  4. [Section 5.1] The paper says LPIPS is used 'to measure content preservation performance' but does not specify the image pair (presumably output versus input). Please state the exact protocol, including for the identity distance metric.
  5. [References] Reference [17] is cited for the DDIM sampler, but the cited title is 'Pseudo numerical methods for diffusion models on manifolds'; the correct DDIM reference [30] is listed separately and should be cited in the implementation details.
  6. [General] The paper frequently refers to 'supplementary material' for additional results, but the supplementary is not included with the arXiv submission; please ensure it is available upon publication.

Circularity Check

1 steps flagged · score 4.0 of 10

LPIPS is minimized in the semantic-adapter training loss and then reused as the content-preservation metric, making that table column partly dependent on the training objective; otherwise the derivation is self-contained.

  1. fitted input called prediction [Eq. (6) in Sec. 4.1; Section 5.1 Metrics; Tables 1 and 2]
    "we introduce a cyclic warping consistency loss to reinforce style preservation during the warping process: Lcwc = L_LPIPS(z_s^0, z_s'^w), where z_s'^w is obtained by warping z_s^w using the transposed correlation matrix M^T. ... Additionally, we use LPIPS [44] to measure content preservation performance."

    The LPIPS distance minimized in Eq. (6) is the same perceptual metric later reported in Tables 1 and 2 as the content-preservation score. The semantic adapter is trained to reduce LPIPS on cyclic warping, so the LPIPS column does not provide an independent check of the method's content-preservation claim; it partly reflects optimization of that metric. The exact inputs differ (reference cycle-warp vs. generated output), so this is a partial metric-dependence rather than a strict identity, but it still weakens LPIPS as external evidence.

full rationale

The core pipeline -- frozen SD U-Net plus semantic adapter for dense warping (Eqs. 3-6), AdaIN-Wavelet latent initialization (Eqs. 10-12), and dual-conditional ControlNet/style-adapter diffusion (Eqs. 7-9, 15) -- is trained on its own reconstruction losses and compared against external baselines. Identity distance (ArcFace) and Gram loss are not direct training objectives, and the ablation studies compare variants of the proposed architecture, so the central derivation is not forced by construction. The main circularity signal is the LPIPS reuse described in the step: LPIPS is a training loss (Eq. 6) and later a headline evaluation metric (Tables 1-2). There is also a minor self-citation: the training split and Gram-loss protocol are credited to the authors' own prior work [37], and [37] is used as the strongest baseline; this is normal practice and not load-bearing, but it adds to the self-referential flavor. No uniqueness theorem or ansatz is smuggled in via citation, and the cross-domain generalization claim, while only qualitatively demonstrated, is an empirical premise rather than a circular reduction.

Assumptions & free parameters 5 free parameters · 5 assumptions · 3 invented entities

The central claim rests on standard diffusion components plus several hand-set hyperparameters (gamma, tau, lambda_m, lambda_c, ControlNet scale). The key domain assumptions are that pretrained features align faces across domains and that self-supervised training with the input as both content and style transfers to arbitrary references. No formal verification is provided.

free parameters (5)
  • gamma style interpolation weight = 1.0 (default)
    Hand-set in Section 4.3 and tuned for best stylization (Figure 8); controls latent blend between content and stylized latents and the style-adapter interpolation.
  • tau softmax temperature = 0.01
    Set in Section 4.1 to sharpen the correlation softmax in warping; chosen by hand.
  • lambda_m mask warping loss weight = 10
    Set in Section 4.4 in the total loss L = L_sem + lambda_c L_cwc + lambda_m L_mask.
  • lambda_c cyclic warping loss weight = 1
    Set in Section 4.4; weights the LPIPS-based cyclic consistency of warping.
  • ControlNet conditional scale = 0.4 for sketch style
    Adjusted at inference (Section 5.3, Figure 11) for sketch stylization; otherwise a default scale is used.
assumptions (5)
  • domain assumption Pretrained Stable Diffusion and CLIP features encode transferable semantic correspondences across image domains.
    Invoked in Section 4.1 for correspondence extraction; no new evidence is provided for cross-domain feature reliability beyond qualitative examples.
  • domain assumption CelebAMask-HQ semantic masks are accurate enough to supervise mask warping and region-specific transfer.
    Masks are used in L_mask (Eq. 5) and for region-controlled results (Figure 7); errors in masks would propagate to the adapter.
  • domain assumption Training with the input image as both content and style guidance (Eq. 15) transfers to arbitrary style references at inference.
    Because no paired style-transfer data exists, stage-2 training conditions on the same image for c_cnt and c_sty; the paper does not analyze the domain shift between self-conditioning and reference conditioning.
  • standard math Haar wavelet transform is orthonormal and invertible for the proposed latent blending.
    Used in Eqs. (7-8) and Eq. (11); standard signal-processing result.
  • ad hoc to paper DDIM inversion provides a latent whose low-frequency components carry the reference image's color tone.
    Section 4.3 relies on the informal assumption that the warped-reference DDIM latent improves color transfer and that wavelet blending fixes blur; no formal justification is given.
invented entities (3)
  • Semantic Adapter
    purpose: Learns to refine dense semantic correspondence between input and reference portraits by combining frozen SD U-Net features and CLIP features with mask and cyclic warping losses.
    Introduced in Section 4.1; the only evidence is in-paper ablation (Figure 4) and downstream metrics, with no external benchmark or theoretical guarantee.
  • Style Adapter
    purpose: Injects reference style into decoupled cross-attention as image-prompt embeddings derived from CLIP and a fixed text prompt.
    Section 4.2, Eq. (9); reuse of IP-Adapter-style architecture, with only in-paper ablation (Figure 5, Table 1).
  • AdaIN-Wavelet Transform
    purpose: Initial latent construction blending low-frequency warped-reference color with high-frequency input structure.
    Section 4.3, Eqs. (10-12); central novel component, supported only by internal ablation (Figure 6, Table 1).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Domain Generalizable Portrait Style Transfer." pith.science (2026). https://pith.science/paper/7VKTOJ37

@misc{pith2026250704243,
  author       = {Pith},
  title        = {Pith review of: Domain Generalizable Portrait Style Transfer},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7VKTOJ37}},
  note         = {Machine review of arXiv:2507.04243}
}
read the original abstract

This paper presents a portrait style transfer method that generalizes well to various different domains while enabling high-quality semantic-aligned stylization on regions including hair, eyes, eyelashes, skins, lips, and background. To this end, we propose to establish dense semantic correspondence between the given input and reference portraits based on a pre-trained model and a semantic adapter, with which we obtain a warped reference semantically aligned with the input. To ensure effective yet controllable style transfer, we devise an AdaIN-Wavelet transform to balance content preservation and stylization by blending low-frequency information of the warped reference with high-frequency information of the input in the latent space. A style adapter is also designed to provide style guidance from the warped reference. With the stylized latent from AdaIN-Wavelet transform, we employ a dual-conditional diffusion model that integrates a ControlNet recording high-frequency information and the style guidance to generate the final result. Extensive experiments demonstrate the superiority of our method. Our code and trained model are available at https://github.com/wangxb29/DGPST.

Figures

Figures reproduced from arXiv: 2507.04243 by the authors.

Figure 1
Figure 1. Domain generalizable portrait style transfer. By training on the CelebAMask-HQ dataset with only 30K portrait photos, our method allows for high-quality semantic-aware style transfer between any two portraits from a wide variety of domains including photo, cartoon, sketch, animation, etc. Abstract This paper presents a portrait style transfer method that generalizes well to various different domains while enabling h… view at source ↗
Figure 2
Figure 2. Framework overview. (Left) Illustration of the proposed portrait style transfer method. We first feed the input and reference portraits into Stable Diffusion and the Semantic Adapter. Next, we extract diffusion features and compute a correlation matrix, which is used to warp the reference portrait. During sampling, we provide structure guidance by extracting the high-frequency information of the input portrait using… view at source ↗
Figure 3
Figure 3. Qualitative comparison with SOTA portrait style transfer methods. spondence learned from diffusion models to obtain a high￾quality warped style reference z s w 0 (Section 4.1). To gen￾erate a realistic portrait, our framework incorporates a dual￾conditional diffusion model (Section 4.2). This model uti￾lizes the ControlNet ϵθ, which extracts high-frequency in￾formation from the content image z c 0 to provide content… view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Effect of our semantic correspondence. For a query point (yellow point in input), the first row shows the similarity map computed by different methods. The second row displays the warped references. The third row shows the outputs generated using these different warped…
Figure 5
Figure 5. Figure 5: Effect of the ControlNet and style adapter. Reference Input Input latent (z c T ) w.r. latent (z s w T ) AdaIN (z cs′ T ) Ours (z cs T ) [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Effect of the initialization of latent noise. “w.r.” stands for warped reference. Reference Input Hair face lips [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 8
Figure 8. Figure 8: Style interpolation. 5. Experiments 5.1. Datasets and Evaluation Metrics Datasets. We evaluate our method on CelebAMask-HQ [13], FFHQ [11], and AAHQ [18]. CelebAMask-HQ con￾tains 30K high-resolution portrait images, each paired with a facial segmentation mask covering …
Figure 9
Figure 9. Figure 9: Results of adding color to grays-scale and sketch portraits. Input Output Input Output Input Output [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]
Figure 10
Figure 10. Figure 10: Results of modernizing old photographs. Reference Output 1 Output 2 Output 3 [PITH_FULL_IMAGE:figures/full_fig_p008_10.png]
Figure 11
Figure 11. Figure 11: Results of transferring real portraits to sketch style. ground. Additionally, our method demonstrates superior performance in content preservation, unlike artistic meth￾ods [5, 41, 43], which often alter facial identity. Please see the supplementary material for more …

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

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

  1. Pretrained Diffusion Models Are Inherently Skipped-Step Samplers

    cs.CV 2025-08 conditional novelty 3.0 of 10

    A DDPM-trained noise predictor can denoise across several time steps in one update because the multi-step posterior is Gaussian and uses the same network.

Reference graph

Works this paper leans on

47 extracted references · 36 canonical work pages · cited by 1 Pith paper

  1. [1]

    Deepfaceediting: deep face generation and editing with disentangled geome- try and appearance control.ACM Transactions on Graphics (TOG), 40(4):1–15, 2021

    Shu-Yu Chen, Feng-Lin Liu, Yu-Kun Lai, Paul L Rosin, Chunpeng Li, Hongbo Fu, and Lin Gao. Deepfaceediting: deep face generation and editing with disentangled geome- try and appearance control.ACM Transactions on Graphics (TOG), 40(4):1–15, 2021. 2, 3, 6, 7

    work page 2021

  2. [2]

    One-shot structure-aware stylized image synthesis

    Hansam Cho, Jonghyun Lee, Seunggyu Chang, and Yonghyun Jeong. One-shot structure-aware stylized image synthesis. InCVPR, pages 8302–8311, 2024. 2

    work page 2024

  3. [3]

    Style injec- tion in diffusion: A training-free approach for adapting large- scale diffusion models for style transfer

    Jiwoo Chung, Sangeek Hyun, and Jae-Pil Heo. Style injec- tion in diffusion: A training-free approach for adapting large- scale diffusion models for style transfer. InCVPR, 2024. 2, 6, 7, 8

    work page 2024

  4. [4]

    Arcface: Additive angular margin loss for deep face recognition

    Jiankang Deng, Jia Guo, Niannan Xue, and Stefanos Zafeiriou. Arcface: Additive angular margin loss for deep face recognition. InCVPR, pages 4690–4699, 2019. 7

    2019

  5. [5]

    z∗: Zero-shot style transfer via attention rearrangement

    Yingying Deng, Xiangyu He, Fan Tang, and Weiming Dong. z∗: Zero-shot style transfer via attention rearrangement. In CVPR, 2024. 2, 6, 7, 8

    work page 2024

  6. [6]

    Gatys, Alexander S

    Leon A. Gatys, Alexander S. Ecker, and Matthias Bethge. Image style transfer using convolutional neural networks. In CVPR, 2016. 2

    work page 2016

  7. [7]

    Mask-guided portrait editing with con- ditional gans

    Shuyang Gu, Jianmin Bao, Hao Yang, Dong Chen, Fang Wen, and Lu Yuan. Mask-guided portrait editing with con- ditional gans. InCVPR, 2019. 2, 3

    work page 2019

  8. [8]

    A robust deep style transfer for headshot portraits.Neurocomputing, 361:164–172, 2019

    Meiqin Guo and Jianmin Jiang. A robust deep style transfer for headshot portraits.Neurocomputing, 361:164–172, 2019. 2

    work page 2019

Show all 47 references
  1. [9]

    Denoising dif- fusion probabilistic models.NeuIPS, 33:6840–6851, 2020

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising dif- fusion probabilistic models.NeuIPS, 33:6840–6851, 2020. 3

    2020

  2. [10]

    Training- free content injection using h-space in diffusion models

    Jaeseok Jeong, Mingi Kwon, and Youngjung Uh. Training- free content injection using h-space in diffusion models. In WACV, pages 5151–5161, 2024. 2

    2024

  3. [11]

    A style-based generator architecture for generative adversarial networks

    Tero Karras, Samuli Laine, and Timo Aila. A style-based generator architecture for generative adversarial networks. In CVPR, 2019. 3, 7

    2019

  4. [12]

    Analyzing and improving the image quality of StyleGAN

    Tero Karras, Samuli Laine, Miika Aittala, Janne Hellsten, Jaakko Lehtinen, and Timo Aila. Analyzing and improving the image quality of StyleGAN. InCVPR, 2020. 3

    2020

  5. [13]

    Maskgan: Towards diverse and interactive facial image ma- nipulation

    Cheng-Han Lee, Ziwei Liu, Lingyun Wu, and Ping Luo. Maskgan: Towards diverse and interactive facial image ma- nipulation. InCVPR, 2020. 7

    2020

  6. [14]

    When stylegan meets stable diffusion: aW + adapter for person- alized image generation

    Xiaoming Li, Xinyu Hou, and Chen Change Loy. When stylegan meets stable diffusion: aW + adapter for person- alized image generation. InCVPR, 2024. 7

    2024

  7. [15]

    Sd4match: Learning to prompt stable diffu- sion model for semantic matching

    Xinghui Li, Jingyi Lu, Kai Han, and Victor Adrian Prisacariu. Sd4match: Learning to prompt stable diffu- sion model for semantic matching. InCVPR, pages 27558– 27568, 2024. 2

    2024

  8. [16]

    Ctrl-x: Controlling structure and appear- ance for text-to-image generation without guidance.arXiv preprint arXiv:2406.07540, 2024

    Kuan Heng Lin, Sicheng Mo, Ben Klingher, Fangzhou Mu, and Bolei Zhou. Ctrl-x: Controlling structure and appear- ance for text-to-image generation without guidance.arXiv preprint arXiv:2406.07540, 2024. 2

    2024 arXiv

  9. [17]

    Pseudo numerical methods for diffusion models on manifolds.arXiv preprint arXiv:2202.09778, 2022

    Luping Liu, Yi Ren, Zhijie Lin, and Zhou Zhao. Pseudo numerical methods for diffusion models on manifolds.arXiv preprint arXiv:2202.09778, 2022. 6

    2022 arXiv

  10. [18]

    Blendgan: Implicitly gan blending for arbitrary stylized face generation

    Mingcong Liu, Qiang Li, Zekui Qin, Guoxin Zhang, Pengfei Wan, and Wen Zheng. Blendgan: Implicitly gan blending for arbitrary stylized face generation. InNeurIPS, 2021. 7

    2021

  11. [19]

    Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101, 2017

    I Loshchilov. Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101, 2017. 6

    2017 arXiv

  12. [20]

    Deep photo style transfer

    Fujun Luan, Sylvain Paris, Eli Shechtman, and Kavita Bala. Deep photo style transfer. InCVPR, 2017. 2

    2017

  13. [21]

    Coordgan: Self-supervised dense correspondences emerge from gans

    Jiteng Mu, Shalini De Mello, Zhiding Yu, Nuno Vasconce- los, Xiaolong Wang, Jan Kautz, and Sifei Liu. Coordgan: Self-supervised dense correspondences emerge from gans. In CVPR, 2022. 7

    2022

  14. [22]

    Pytorch: An imperative style, high-performance deep learning library

    Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zem- ing Lin, Natalia Gimelshein, Luca Antiga, et al. Pytorch: An imperative style, high-performance deep learning library. NeurIPS, 32, 2019. 6

    2019

  15. [23]

    Dead- iff: An efficient stylization diffusion model with disentan- gled representations

    Tianhao Qi, Shancheng Fang, Yanze Wu, Hongtao Xie, Ji- awei Liu, Lang Chen, Qian He, and Yongdong Zhang. Dead- iff: An efficient stylization diffusion model with disentan- gled representations. InCVPR, pages 8693–8702, 2024. 2

    2024

  16. [24]

    Learn- ing transferable visual models from natural language super- vision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learn- ing transferable visual models from natural language super- vision. InICLR, pages 8748–8763, 2021. 3

    2021

  17. [25]

    High-resolution image syn- thesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj¨orn Ommer. High-resolution image syn- thesis with latent diffusion models. InCVPR, pages 10684– 10695, 2022. 3

    2022

  18. [26]

    Paint- ing style transfer for head portraits using convolutional neu- 9 ral networks.ACM Transactions on Graphics (ToG), 35(4): 1–18, 2016

    Ahmed Selim, Mohamed Elgharib, and Linda Doyle. Paint- ing style transfer for head portraits using convolutional neu- 9 ral networks.ACM Transactions on Graphics (ToG), 35(4): 1–18, 2016. 2

    2016

  19. [27]

    Freeman, and Fr´edo Durand

    Yichang Shih, Sylvain Paris, Connelly Barnes, William T. Freeman, and Fr´edo Durand. Style transfer for headshot por- traits.ACM Transactions on Graphics (TOG), 33:1 – 14,

  20. [28]

    Portrait lighting trans- fer using a mass transport approach.ACM Trans

    Zhixin Shu, Sunil Hadap, Eli Shechtman, Kalyan Sunkavalli, Sylvain Paris, and Dimitris Samaras. Portrait lighting trans- fer using a mass transport approach.ACM Trans. Graph., 36 (4), 2017. 2

    2017

  21. [29]

    Agilegan: Stylizing portraits by inversion-consistent transfer learning

    Guoxian Song, Linjie Luo, Jing Liu, Wan-Chun Ma, Chun- pong Lai, Chuanxia Zheng, and Tat-Jen Cham. Agilegan: Stylizing portraits by inversion-consistent transfer learning. ACM Trans. Graph., 40(4), 2021. 3

    2021

  22. [30]

    Denoising diffusion implicit models.arXiv preprint arXiv:2010.02502, 2020

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models.arXiv preprint arXiv:2010.02502, 2020. 3

    2010 arXiv

  23. [31]

    Ide-3d: Interactive disentangled edit- ing for high-resolution 3d-aware portrait synthesis.ACM Transactions on Graphics (TOG), 41(6):1–10, 2022

    Jingxiang Sun, Xuan Wang, Yichun Shi, Lizhen Wang, Jue Wang, and Yebin Liu. Ide-3d: Interactive disentangled edit- ing for high-resolution 3d-aware portrait synthesis.ACM Transactions on Graphics (TOG), 41(6):1–10, 2022. 2

    2022

  24. [32]

    Emergent correspondence from image diffusion.NeurIPS, 36:1363–1389, 2023

    Luming Tang, Menglin Jia, Qianqian Wang, Cheng Perng Phoo, and Bharath Hariharan. Emergent correspondence from image diffusion.NeurIPS, 36:1363–1389, 2023. 2, 4, 5

    2023

  25. [33]

    Towards harmonized regional style transfer and ma- nipulation for facial images.Computational Visual Media, 9 (2):351–366, 2023

    Cong Wang, Fan Tang, Yong Zhang, Tieru Wu, and Weiming Dong. Towards harmonized regional style transfer and ma- nipulation for facial images.Computational Visual Media, 9 (2):351–366, 2023. 2, 3

    2023

  26. [34]

    Instantstyle: Free lunch towards style- preserving in text-to-image generation.arXiv preprint arXiv:2404.02733, 2024

    Haofan Wang, Qixun Wang, Xu Bai, Zekui Qin, and Anthony Chen. Instantstyle: Free lunch towards style- preserving in text-to-image generation.arXiv preprint arXiv:2404.02733, 2024. 2

    2024 arXiv

  27. [35]

    Instantstyle-plus: Style transfer with content-preserving in text-to-image generation.arXiv preprint arXiv:2407.00788, 2024

    Haofan Wang, Peng Xing, Renyuan Huang, Hao Ai, Qixun Wang, and Xu Bai. Instantstyle-plus: Style transfer with content-preserving in text-to-image generation.arXiv preprint arXiv:2407.00788, 2024. 2, 5, 6, 7

    2024 arXiv

  28. [36]

    Instantid: Zero-shot identity-preserving gener- ation in seconds.arXiv preprint arXiv:2401.07519, 2024

    Qixun Wang, Xu Bai, Haofan Wang, Zekui Qin, and An- thony Chen. Instantid: Zero-shot identity-preserving gener- ation in seconds.arXiv preprint arXiv:2401.07519, 2024. 2

    2024 arXiv

  29. [37]

    Towards photorealistic portrait style transfer in un- constrained conditions.IEEE Transactions on Visualization and Computer Graphics, pages 1–14, 2025

    Xinbo Wang, Qing Zhang, Yongwei Nie, and Wei-Shi Zheng. Towards photorealistic portrait style transfer in un- constrained conditions.IEEE Transactions on Visualization and Computer Graphics, pages 1–14, 2025. 2, 6, 7

    2025

  30. [38]

    Stylediffusion: Controllable disentangled style transfer via diffusion models

    Zhizhong Wang, Lei Zhao, and Wei Xing. Stylediffusion: Controllable disentangled style transfer via diffusion models. InICCV, pages 7677–7689, 2023. 2

    2023

  31. [39]

    Vtoonify: Controllable high-resolution portrait video style transfer.ACM Trans

    Shuai Yang, Liming Jiang, Ziwei Liu, and Chen Change Loy. Vtoonify: Controllable high-resolution portrait video style transfer.ACM Trans. Graph., 41(6), 2022. 3

    2022

  32. [40]

    Pastiche master: Exemplar-based high-resolution por- trait style transfer

    Shuai Yang, Liming Jiang, Ziwei Liu, and Chen Change Loy. Pastiche master: Exemplar-based high-resolution por- trait style transfer. InCVPR, 2022. 2, 3

    2022

  33. [41]

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

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

    arXiv

  34. [42]

    A tale of two features: Stable diffusion complements dino for zero-shot semantic correspondence.NeurIPS, 36,

    Junyi Zhang, Charles Herrmann, Junhwa Hur, Luisa Pola- nia Cabrera, Varun Jampani, Deqing Sun, and Ming-Hsuan Yang. A tale of two features: Stable diffusion complements dino for zero-shot semantic correspondence.NeurIPS, 36,

  35. [43]

    Adding conditional control to text-to-image diffusion models

    Lvmin Zhang, Anyi Rao, and Maneesh Agrawala. Adding conditional control to text-to-image diffusion models. In ICCV, pages 3836–3847, 2023. 6, 7, 8

    2023

  36. [44]

    Efros, Eli Shecht- man, and Oliver Wang

    Richard Zhang, Phillip Isola, Alexei A. Efros, Eli Shecht- man, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. InCVPR, 2018. 7

    2018

  37. [45]

    Prospect: Prompt spectrum for attribute-aware personalization of diffusion models.ACM Transactions on Graphics (TOG), 42(6):1–14, 2023

    Yuxin Zhang, Weiming Dong, Fan Tang, Nisha Huang, Haibin Huang, Chongyang Ma, Tong-Yee Lee, Oliver Deussen, and Changsheng Xu. Prospect: Prompt spectrum for attribute-aware personalization of diffusion models.ACM Transactions on Graphics (TOG), 42(6):1–14, 2023. 2

    2023

  38. [46]

    Inversion-based style transfer with diffusion models

    Yuxin Zhang, Nisha Huang, Fan Tang, Haibin Huang, Chongyang Ma, Weiming Dong, and Changsheng Xu. Inversion-based style transfer with diffusion models. In CVPR, pages 10146–10156, 2023. 2

    2023

  39. [47]

    Deformable one- shot face stylization via dino semantic guidance

    Yang Zhou, Zichong Chen, and Hui Huang. Deformable one- shot face stylization via dino semantic guidance. InCVPR, pages 7787–7796, 2024. 2, 3, 6, 7 10

    2024

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.