Pith. sign in

REVIEW 5 major objections 6 minor 26 references

One-Shot Multilingual Font Generation Via ViT

T0 review · 5 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

Pith's one-line read A ViT-MAE model with cross-attention can generate fonts for unseen and user-drawn characters in one shot.

desk verdict A plausible ViT/MAE recipe for one-shot font transfer whose headline generalization claim rests on a single unquantified example and is undercut by its own tables; worth a skim, not a citation. read the letter →

arxiv 2412.11342 v1 pith:B3SSB3NA submitted 2024-12-15 cs.CV

classification cs.CV
keywords fontgenerationone-shotlearningVisionTransformermaskedautoencodercross-attentionmultilingualglyphsynthesisretrieval-augmented
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 tries to establish that a single Vision Transformer model, pretrained with Masked Autoencoding, can generate fonts for Chinese, Japanese, Korean, and English in a one-shot setting: given one content character and one style sample, it can render the content in the target style even when the character is absent from every reference library, including hand-drawn invented glyphs. If true, this would remove the need to hand-craft or collect each character in each style, a bottleneck that makes logographic font design expensive. The method's central mechanism is cross-attention fusion of content and style embeddings from two MAE-pretrained ViT encoders, followed by a decoder and a combined perceptual-plus-pixel loss. The paper argues for the claim with human ratings of style transfer, qualitative cross-language examples, invented-character demos, and numerical metrics on four unseen-font settings. A retrieval module is introduced to repair specific hard cases such as confusing similar radicals.

What carries the argument

The central mechanism is a cross-attention bi-encoder built on an MAE-pretrained ViT backbone: a content encoder embeds the input glyph, a style encoder embeds the style reference, and the content embedding serves as queries while the style embedding provides keys and values, so the decoder reconstructs a glyph that keeps content structure while adopting style details. The training objective is a combined loss $L_{\text{total}} = \alpha L_{\text{content}} + \beta L_{\text{style}} + \gamma L_{\text{MSE}}$ with perceptual features from a pre-trained VGG19 network, followed by a short L1 refinement phase. A retrieval module uses FAISS nearest-neighbor search over content-encoder embeddings to supply a structurally similar style reference for hard cases.

What would settle it

Generate an invented logogram with no training-set analogue, apply an unseen font style, and ask native readers to identify which font was used; if they perform at chance, or if the output silently turns into a visually similar real character, the one-shot generalization claim fails.

Watch

Extended reading notes

Core claim

The paper claims that a ViT-based bi-encoder system pretrained with Masked Autoencoding performs one-shot multilingual font generation: it takes a content glyph and one style sample, then produces a new glyph that preserves the content and adopts the style, for scripts including Chinese, Japanese, Korean, and English. The authors hold that this works even for content characters that are unseen, unknown, and user-crafted, with no reference character in the style library, because MAE pretraining teaches the encoders robust glyph structure and the cross-attention fusion selectively applies style features. The experiments report style transfer recognized by human raters, cross-language transfer in both directions, and qualitative success on invented handwriting, alongside numerical metrics on four unseen-font test settings.

Load-bearing premise

The load-bearing premise is that a single style sample carries enough information about a font for the cross-attention fusion to redraw any content glyph in that font, even a character never seen during training, and that content and style signals stay cleanly separated inside the model.

Editorial extensions

If this is right

  • A font designer can render an entirely new character, handwritten or invented, in any style using just one style sample, without hunting for a matching reference character.
  • Cross-language style transfer becomes bidirectional, so Chinese content can be rendered in Japanese hiragana style and vice versa, enabling mixed-script typography.
  • The model's reliance on MAE pretraining on font images suggests that new scripts or styles can be added by pretraining on unlabeled glyph images rather than collecting large paired datasets.
  • The retrieval module, though neutral on aggregate metrics, demonstrably fixes specific hard cases such as confusing visually similar radicals, which matters in practice more than average scores.
  • The same pipeline could be adapted to other single-object image-to-image style transfer tasks beyond fonts, such as icons or symbols.

Reading between the lines

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

  • The RAG results imply that retrieval is best understood as a targeted repair mechanism, not a quality booster: improving it may require measuring failure-case frequency, not global FID or SSIM.
  • Because the style encoder can be invoked multiple times and averaged, the paper's untested few-shot extension is a natural next experiment; it would test whether cross-attention fusion degrades or improves when multiple style keys are supplied.
  • The claim of handling unknown characters implies a testable disentanglement property: content embeddings should be invariant to style and style embeddings invariant to content. That property is not directly measured here, and an experiment probing it would sharpen the claim.
  • The model's tolerance for hand-drawn invented inputs suggests a path toward interactive font design tools, but the authors note the handwriting investigation was limited, so verifying style consistency across many writers would be needed before deployment.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 6 minor

Summary. The paper proposes a ViT-based, MAE-pretrained model for one-shot multilingual font generation. It uses two encoders (content and style), a cross-attention fusion module, and a decoder, trained with a combination of VGG-based perceptual losses and MSE, followed by an L1 refinement. A retrieval-augmented guidance (RAG) module is added to retrieve style references from a known inventory. The authors claim the model generates high-quality fonts for unseen, unknown, and user-crafted characters across Chinese, Japanese, Korean, and English without requiring a reference character in its library. Experiments include qualitative examples, a small human evaluation, and quantitative comparisons on L1/RMSE, SSIM, LPIPS, and FID across several unseen-font/character settings.

Significance. If substantiated, the paper would address a real need: one-shot, library-free, cross-lingual font generation, including for arbitrary or invented characters. The use of MAE-pretrained ViTs and a cross-attention bi-encoder is a plausible design direction, and the paper demonstrates that the architecture can produce recognizably style-transferred glyphs in several languages. However, the significance is currently only potential; the quantitative evidence contradicts the 'high-quality' claim, the RAG module is shown to hurt all metrics, and the flagship user-crafted-character result rests on a single unquantified example. The novelty of the architecture is incremental relative to existing transformer-based font generation methods, and the claimed library-free generalization is unsupported by the experimental design.

major comments (5)
  1. [§4.4 and §6.2.3] The central one-shot generalization claim depends on the assumption that the content encoder and style encoder produce disentangled representations and that cross-attention (content queries attending to style keys/values) can inject style without corrupting content for arbitrary novel glyphs. The paper provides no probe, ablation, or embedding-space analysis to validate this factorization. Because Eq. (1) is supervised only where ground truth exists, the model's behavior on user-crafted characters (Figure 9) is an unverified extrapolation from the trained regime. This is a load-bearing gap for the paper's headline claim.
  2. [Tables 1 and 3] The quantitative results undermine the 'high-quality' claim in the Abstract. In the SS (style font unseen) setting, the proposed method achieves L1 0.187 vs. 0.078 (DG-Font) and 0.074 (CF-Font), SSIM 0.666 vs. 0.685 and 0.701, and LPIPS 0.192 vs. 0.120 and 0.118; only FID is better (25.56 vs. 27.98 and 26.51). No error bars or significance tests are reported. The discussion in §6.2.4, appealing to shorter training, one-shot vs. few-shot, and dataset difficulty, is post-hoc and does not provide a controlled comparison.
  3. [Table 2 and §6.2.5] The RAG module makes every reported metric worse in every unseen setting (e.g., SS L1 increases from 0.187 to 0.199, SSIM falls from 0.666 to 0.654, LPIPS rises from 0.192 to 0.203). This directly contradicts the RAG hypothesis in §4.4 that 'characters with similar content or structure can provide additional style information.' The positive evidence is limited to three hand-picked examples (Figures 10–12) with no quantitative support. The conclusion that RAG 'plays a vital role in practical applications' is a post-hoc qualitative assertion rather than a demonstrated result.
  4. [Figure 9 and §7] The claim of generating 'user-crafted characters' (Abstract and §6.2.3) rests on a single handwritten, invented example with no ground truth, no human rating, and no recognition test. Section 7 explicitly concedes that 'large-scale investigation was not conducted.' This is insufficient evidence for one of the paper's headline capabilities.
  5. [§6.2.4] The comparison with DG-Font and CF-Font is not a controlled head-to-head. The authors acknowledge that those baselines were trained for 20k iterations versus their 10, use few-shot rather than one-shot settings, and are evaluated on datasets of different difficulty. Consequently, Table 3 does not support the statement that 'our model's results are not far behind the current state of the art.' A fair comparison with matched training budgets, evaluation protocols, and dataset characteristics is needed before any comparative claim can be made.
minor comments (6)
  1. [§4.1] A patch size of 16 on a 24×24 input is geometrically inconsistent since 24 is not divisible by 16. Please specify whether padding, overlapping patches, or a different resolution is actually used.
  2. [§3] The dataset statistics are internally inconsistent: the text mentions '800k images for pretrain' but later 'around 1M for training'; it also says 'total of 154 styles' while the dataset is said to have 308 styles. Please clarify the exact numbers.
  3. [Tables 1 and 3] Table 1 reports MSE, while Table 3 reports RMSE. To allow direct comparison, the same metric (with the same definition) should be used in both tables.
  4. [§6.2.1] The human evaluation is described with insufficient detail: the number of trials, the stimuli selection, the instructions, and inter-rater agreement are not reported. Six subjects is a small sample, and the conclusion that all Group 1 participants rated the transfer as '2' should be treated cautiously.
  5. [§4.4] The text states the RAG module uses 'IndexFlatL2' but 'performs similarity searches based on cosine similarity.' FAISS's IndexFlatL2 uses L2 distance, not cosine similarity; please clarify which metric is actually used.
  6. [§6.2.4] The statement that Diffusion-Font 'cannot consistently perform valid style transfer' is made without supporting evidence or experimental details. Either provide the empirical results or remove the claim.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the pipeline is a supervised encoder–decoder trained on paired data; the unverified generalization claims are evidentiary, not circular.

full rationale

Walking the derivation chain: the model is trained with a supervised combined loss (Eq. 1) against ground-truth glyphs, with loss weights and mask ratio chosen by experiment; none of these hyperparameters are renamed predictions. The content/style encoders are MAE-pretrained on the dataset and then fine-tuned, and the baselines are trained on the same data, so the comparisons are not circular. The RAG module builds a FAISS index from the content encoder's own embeddings and retrieves from that same index; this is self-referential in the sense that the retrieval criterion is internal to the model, but it is not a case of fitting a parameter and then predicting that same parameter, and the paper's own Table 2 reports that RAG does not improve the standard metrics, so the retrieval claim is not being forced by construction. The central claim about unseen, unknown, and user-crafted characters rests on qualitative examples (Fig. 9) and on the unverified assumption that content and style embeddings disentangle; Section 7 concedes the handwriting investigation was not large-scale. Those are evidentiary weaknesses or correctness risks, not circularity: no equation reduces to its inputs, no fitted value is relabeled as a prediction, and no load-bearing argument is justified by self-citation (the reference list contains no self-citations). Accordingly the circularity score is 0.

Assumptions & free parameters 7 free parameters · 6 assumptions · 0 invented entities

The paper's inventions are design choices (loss weights, mask ratio, layer selections) and a method module (RAG), not new physical or mathematical entities. The central claim rests on domain assumptions about content/style separability and on MAE pretraining being beneficial, neither of which is rigorously demonstrated. The RAG hypothesis, in particular, is contradicted by the paper's own quantitative results.

free parameters (7)
  • Content loss weight α = 0.1
    Chosen by hand to balance raw scales of perceptual losses; Section 4.4.
  • Style loss weight β = 0.4
    Set higher than α because the primary objective is style transfer; Section 4.4.
  • MSE loss weight γ = 1.0
    Increased to keep a pixel-level anchor for edges and glyph boundaries; Section 4.4.
  • MAE mask ratio = 0.65
    Reduced from 0.75 after experiments showed poor reconstruction on complex glyphs; Section 4.1 and Figure 2.
  • Input image size = 24x24
    Reduced from 80x80 because the model does not learn well at the larger size; Section 4.1.
  • Patch size = 16
    Architectural choice; at 24x24 this yields only a 2x2 patch grid; Section 4.1.
  • VGG19 layer selection for content and style losses
    Hand-selected layers (relu22 for content; relu11, relu21, relu31, relu41, relu51 for style) in Section 4.4.
assumptions (6)
  • standard math Neural network training with backpropagation minimizes the stated losses.
    Standard machine learning assumption implicit throughout the experiments.
  • domain assumption Each glyph can be treated as a single-object image of fixed position and size, making self-attention suitable.
    Justifies using ViT with fixed-size inputs; stated in Section 4.2.
  • domain assumption A character's visual identity consists of separable content (glyph identity) and style (stroke thickness, curvature, texture), and the encoders can extract these.
    Core assumption of the bi-encoder architecture; Sections 4.2 and 4.4.
  • ad hoc to paper MAE pretraining on the font dataset improves downstream font generation compared to training from scratch or using a generic pretrained MAE.
    Motivates the method; only reconstruction examples in Section 4.1 are shown, not a final-task ablation.
  • ad hoc to paper Reducing the MAE mask ratio from 0.75 to 0.65 improves the learned representations.
    Based on the reconstruction comparison in Figure 2, not on final generation metrics.
  • ad hoc to paper The RAG hypothesis: characters with similar content or structure provide additional useful style information.
    Load-bearing for the RAG claims; Section 4.4. Table 2 shows no metric improvement, so this assumption is not supported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of One-Shot Multilingual Font Generation Via ViT." pith.science (2026). https://pith.science/paper/B3SSB3NA

@misc{pith2026241211342,
  author       = {Pith},
  title        = {Pith review of: One-Shot Multilingual Font Generation Via ViT},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/B3SSB3NA}},
  note         = {Machine review of arXiv:2412.11342}
}
read the original abstract

Font design poses unique challenges for logographic languages like Chinese, Japanese, and Korean (CJK), where thousands of unique characters must be individually crafted. This paper introduces a novel Vision Transformer (ViT)-based model for multi-language font generation, effectively addressing the complexities of both logographic and alphabetic scripts. By leveraging ViT and pretraining with a strong visual pretext task (Masked Autoencoding, MAE), our model eliminates the need for complex design components in prior frameworks while achieving comprehensive results with enhanced generalizability. Remarkably, it can generate high-quality fonts across multiple languages for unseen, unknown, and even user-crafted characters. Additionally, we integrate a Retrieval-Augmented Guidance (RAG) module to dynamically retrieve and adapt style references, improving scalability and real-world applicability. We evaluated our approach in various font generation tasks, demonstrating its effectiveness, adaptability, and scalability.

Figures

Figures reproduced from arXiv: 2412.11342 by the authors.

Figure 1
Figure 1. From top to bottom are Chinese, Japanese, Korean and [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The comparison showed that it is necessary to pretrain [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Our proposed model utilizes a cross-attention mecha [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (8 more)
Figure 5
Figure 5. Figure 5: In each row, the first image is the content image, the [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 4
Figure 4. Figure 4: DiffuserFont captures content but deviates significantly [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 7
Figure 7. Figure 7: The input content is Chinese, and the style input [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]
Figure 8
Figure 8. Figure 8: Subfigures (a) and (b) show the same Japanese hiragana [PITH_FULL_IMAGE:figures/full_fig_p006_8.png]
Figure 9
Figure 9. Figure 9: The input content is a made-up handwritten word, and [PITH_FULL_IMAGE:figures/full_fig_p006_9.png]
Figure 10
Figure 10. Figure 10: The target character includes the three-dot ”water” rad [PITH_FULL_IMAGE:figures/full_fig_p007_10.png]
Figure 11
Figure 11. Figure 11: The RAG module resolves errors in generating the [PITH_FULL_IMAGE:figures/full_fig_p007_11.png]
Figure 12
Figure 12. Figure 12: For the challenging task of generating hiragana char [PITH_FULL_IMAGE:figures/full_fig_p007_12.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

26 extracted references · 16 canonical work pages

  1. [3]

    Multi- Content GAN for Few-Shot Font Style Transfer, 2017

    Samaneh Azadi, Matthew Fisher, Vladimir Kim, Zhaowen Wang, Eli Shechtman, and Trevor Darrell. Multi- Content GAN for Few-Shot Font Style Transfer, 2017. arXiv:1712.00516 [cs]. 1

  2. [4]

    BEiT: BERT Pre-Training of Image Transformers, 2022

    Hangbo Bao, Li Dong, Songhao Piao, and Furu Wei. BEiT: BERT Pre-Training of Image Transformers, 2022. arXiv:2106.08254. 2

  3. [5]

    Rae, Erich Elsen, and Laurent Sifre

    Sebastian Borgeaud, Arthur Mensch, Jordan Hoffmann, Trevor Cai, Eliza Rutherford, Katie Millican, George van den Driessche, Jean-Baptiste Lespiau, Bogdan Damoc, Aidan Clark, Diego de Las Casas, Aurelia Guy, Jacob Menick, Ro- man Ring, Tom Hennigan, Saffron Huang, Loren Maggiore, Chris Jones, Albin Cassirer, Andy Brock, Michela Paganini, Geoffrey Irving, O...

  4. [6]

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale, 2021

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, Jakob Uszkoreit, and Neil Houlsby. An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale, 2021. arXiv:2010.11929 [cs]. 1, 2

  5. [7]

    ScrabbleGAN: Semi-Supervised Varying Length Handwritten Text Generation, 2020

    Sharon Fogel, Hadar Averbuch-Elor, Sarel Cohen, Shai Ma- zor, and Roee Litman. ScrabbleGAN: Semi-Supervised Varying Length Handwritten Text Generation, 2020. arXiv:2003.10557 [cs]. 1

  6. [8]

    GlyphGAN: Style-Consistent Font Generation Based on Generative Adversarial Networks

    Hideaki Hayashi, Kohtaro Abe, and Seiichi Uchida. Glyph- GAN: Style-Consistent Font Generation Based on Genera- tive Adversarial Networks, 2019. arXiv:1905.12502 [cs]. 1

  7. [9]

    Diff-Font: Diffusion Model for Robust One-Shot Font Generation

    Haibin He, Xinyuan Chen, Chaoyue Wang, Juhua Liu, Bo Du, Dacheng Tao, and Qiao Yu. Diff-Font: Diffusion Model for Robust One-Shot Font Generation. International Journal of Computer Vision, 2024. 1, 2

  8. [10]

    Masked Autoencoders Are Scal- able Vision Learners, 2021

    Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Doll´ar, and Ross Girshick. Masked Autoencoders Are Scal- able Vision Learners, 2021. arXiv:2111.06377 version: 2. 1, 2

Show all 26 references
  1. [11]

    Ar- bitrary Font Generation by Encoder Learning of Disentan- gled Features

    Jeong-Sik Lee, Rock-Hyun Baek, and Hyun-Chul Choi. Ar- bitrary Font Generation by Encoder Learning of Disentan- gled Features. Sensors, 22(6):2374, 2022. Number: 6 Pub- lisher: Multidisciplinary Digital Publishing Institute. 1

  2. [12]

    Few-shot Font Style Transfer between Different Languages

    Chenhao Li, Yuta Taniguchi, Min Lu, and Shin’ichi Konomi. Few-shot Font Style Transfer between Different Languages. In 2021 IEEE Winter Conference on Applications of Com- puter Vision (WACV) , pages 433–442, 2021. ISSN: 2642-

  3. [13]

    FontTransformer: Few-shot High-resolution Chinese Glyph Image Synthesis via Stacked Transformers, 2022

    Yitian Liu and Zhouhui Lian. FontTransformer: Few-shot High-resolution Chinese Glyph Image Synthesis via Stacked Transformers, 2022. arXiv:2210.06301 [cs]. 2

  4. [14]

    Hand- written Text Generation from Visual Archetypes, 2023

    Vittorio Pippi, Silvia Cascianelli, and Rita Cucchiara. Hand- written Text Generation from Visual Archetypes, 2023. arXiv:2303.15269 [cs]. 2

  5. [15]

    Hiera: A Hier- archical Vision Transformer without the Bells-and-Whistles,

    Chaitanya Ryali, Yuan-Ting Hu, Daniel Bolya, Chen Wei, Haoqi Fan, Po-Yao Huang, Vaibhav Aggarwal, Arkabandhu Chowdhury, Omid Poursaeed, Judy Hoffman, Jitendra Ma- lik, Yanghao Li, and Christoph Feichtenhofer. Hiera: A Hier- archical Vision Transformer without the Bells-and-Whistles,

  6. [16]

    Learning to Write Stylized Chinese Characters by Reading a Handful of Examples, 2018

    Danyang Sun, Tongzheng Ren, Chongxun Li, Hang Su, and Jun Zhu. Learning to Write Stylized Chinese Characters by Reading a Handful of Examples, 2018. arXiv:1712.06424 [cs, stat]. 1

  7. [17]

    Few-Shot Font Generation by Learning Fine-Grained Local Styles, 2022

    Licheng Tang, Yiyang Cai, Jiaming Liu, Zhibin Hong, Ming- ming Gong, Minhu Fan, Junyu Han, Jingtuo Liu, Errui Ding, and Jingdong Wang. Few-Shot Font Generation by Learning Fine-Grained Local Styles, 2022. arXiv:2205.09965 [cs]. 2

  8. [18]

    kaonashi-tyc/Rewrite, 2024

    Yuchen Tian. kaonashi-tyc/Rewrite, 2024. original-date: 2016-10-26T03:11:46Z. 1

  9. [19]

    kaonashi-tyc/zi2zi, 2024

    Yuchen Tian. kaonashi-tyc/zi2zi, 2024. original-date: 2017- 02-17T23:18:04Z. 1

  10. [20]

    Gomez, Lukasz Kaiser, and Illia Polosukhin

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszko- reit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention Is All You Need, 2023. arXiv:1706.03762 [cs]. 2

  11. [21]

    Zi- GAN: Fine-grained Chinese Calligraphy Font Generation via a Few-shot Style Transfer Approach

    Qi Wen, Shuang Li, Bingfeng Han, and Yi Yuan. Zi- GAN: Fine-grained Chinese Calligraphy Font Generation via a Few-shot Style Transfer Approach. In Proceedings of the 29th ACM International Conference on Multimedia , pages 621–629, 2021. arXiv:2108.03596 [cs]. 1

  12. [22]

    DG- Font: Deformable Generative Networks for Unsupervised Font Generation, 2021

    Yangchen Xie, Xinyuan Chen, Li Sun, and Yue Lu. DG- Font: Deformable Generative Networks for Unsupervised Font Generation, 2021. arXiv:2104.03064 [cs]. 1, 2

  13. [23]

    MetaScript: Few-Shot Handwritten Chinese Content Generation via Generative Adversarial Networks,

    Xiangyuan Xue, Kailing Wang, Jiazi Bu, Qirui Li, and Zhiyuan Zhang. MetaScript: Few-Shot Handwritten Chinese Content Generation via Generative Adversarial Networks,

  14. [24]

    FontDiffuser: One-Shot Font Generation via Denoising Diffusion with Multi-Scale Content Aggregation and Style Contrastive Learning, 2023

    Zhenhua Yang, Dezhi Peng, Yuxin Kong, Yuyi Zhang, Cong Yao, and Lianwen Jin. FontDiffuser: One-Shot Font Generation via Denoising Diffusion with Multi-Scale Content Aggregation and Style Contrastive Learning, 2023. arXiv:2312.12142 [cs]. 2

  15. [25]

    Few Shot POP Chinese Font Style Transfer using CycleGAN

    Yong Yu. Few Shot POP Chinese Font Style Transfer using CycleGAN. Journal of Physics: Conference Series, 2171(1): 012031, 2022. Publisher: IOP Publishing. 1

  16. [26]

    Sep- arating Style and Content for Generalized Style Transfer,

    Yexun Zhang, Ya Zhang, Wenbin Cai, and Jie Chang. Sep- arating Style and Content for Generalized Style Transfer,

  17. [2018]

    arXiv:1711.06454 [cs]. 1

  18. [2023]

    arXiv:2312.16251 [cs]. 1, 2

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.