pith. machine review for the scientific record. sign in

arxiv: 2605.12138 · v1 · submitted 2026-05-12 · 💻 cs.CV · cs.CL· cs.IR

Recognition: no theorem link

Design Your Ad: Personalized Advertising Image and Text Generation with Unified Autoregressive Models

Yexing Xu , Wei Feng , Shen Zhang , Haohan Wang , Yuxin Qin , Yaoyu Li , Ao Ma , Yuhao Luo
show 10 more authors
Lu Wang Xudong Ren Haoran Wang Run Ling Zheng Zhang Jingjing Lv Junjie Shen Ching Law Longguang Wang Yulan Guo
Authors on Pith no claims yet

Pith reviewed 2026-05-13 07:31 UTC · model grok-4.3

classification 💻 cs.CV cs.CLcs.IR
keywords personalized advertisingautoregressive modelsimage-text generationuser preference modelinge-commerceclick historyunified framework
0
0 comments X

The pith

A single autoregressive model jointly generates personalized advertising images and texts from historical user clicks.

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

The paper proposes Uni-AdGen, a unified autoregressive model that produces both images and texts for advertisements tailored to individual users from their past click data. It replaces separate image and text pipelines that rely on average click-through rates with one framework containing a foreground perception module and a coarse-to-fine preference module to extract personal interests from noisy history. The model uses instruction tuning to improve realism and is supported by a new large-scale dataset and a background similarity metric for training and testing. If the approach succeeds, it simplifies ad creation while matching user tastes more closely than population-level methods.

Core claim

Uni-AdGen employs a single autoregressive framework to produce both advertising images and texts, enhanced by a foreground perception module and instruction tuning for realism, together with a coarse-to-fine preference understanding module that captures user interests from noisy multimodal historical behaviors to drive personalized generation, and extensive experiments show it outperforms baselines on the PAd1M dataset using the Product Background Similarity metric.

What carries the argument

Uni-AdGen: unified autoregressive model with foreground perception module, instruction tuning, and coarse-to-fine preference understanding module that conditions generation on user click history.

If this is right

  • Image and text ads can be generated jointly in one pass while remaining consistent with user-specific interests.
  • Preference extraction from multimodal click history replaces reliance on average CTR signals.
  • The PAd1M dataset enables large-scale supervised training for personalized ad tasks.
  • The Product Background Similarity metric supports direct evaluation of how well generated ads match product contexts.

Where Pith is reading between the lines

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

  • The same unified autoregressive structure could be tested on other multimodal content tasks such as personalized product descriptions or social posts.
  • Streaming click data could be fed into the preference module for near-real-time ad adaptation if latency allows.
  • Direct integration with existing recommender systems would let click signals flow straight into ad generation without intermediate CTR averaging.

Load-bearing premise

Historical click behaviors supply a reliable signal of individual preferences that the coarse-to-fine module extracts without bias from noise.

What would settle it

An A/B test in which ads produced by the model receive no higher click-through rates than those from separate baseline systems would falsify the personalization advantage.

Figures

Figures reproduced from arXiv: 2605.12138 by Ao Ma, Ching Law, Haohan Wang, Haoran Wang, Jingjing Lv, Junjie Shen, Longguang Wang, Lu Wang, Run Ling, Shen Zhang, Wei Feng, Xudong Ren, Yaoyu Li, Yexing Xu, Yuhao Luo, Yulan Guo, Yuxin Qin, Zheng Zhang.

Figure 1
Figure 1. Figure 1: Problems in existing methods and user historical behaviors. (a) Existing methods use separate models for advertisement gener [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: An overview of PAd1M and experimental results of the PBS metric. (a) Visualization of PAd1M, showing the target advertisement [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: An overview of our method. Our framework combines (a) the unified advertisement generative model (Uni-AdGen) with (b) the [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The qualitative results of personalized advertisement generation task. Our method generates images with colors and scenes closer [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: The examples of shape hallucination. atmospheric or blurred backgrounds are exempt from spatial con￾sistency requirements. Rejected Floating products & inconsistent angles. Accepted Atmospheric or blurred backgrounds [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: The examples of space mismatch. Visual perception. This criterion evaluates product distinctive￾ness and potential discomfort. Images containing distorted human body parts (e.g., faces, hands) that cause discomfort or inappropri￾ate associations are rejected, unless bodies naturally interact with products shown in [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: The examples of visual perception. 7.2. Advertising Text Generation Generated advertising texts are evaluated based on factuality and readability as shown in [PITH_FULL_IMAGE:figures/full_fig_p013_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: More visualization examples in the PAd1M dataset. [PITH_FULL_IMAGE:figures/full_fig_p014_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: The examples of human evaluation for advertising [PITH_FULL_IMAGE:figures/full_fig_p014_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: The template pools for advertising text generation task. [PITH_FULL_IMAGE:figures/full_fig_p015_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: The template pools for image prompt generation. [PITH_FULL_IMAGE:figures/full_fig_p015_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: The instruction template used for instruction tuning. [PITH_FULL_IMAGE:figures/full_fig_p017_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: The generated advertising images of different methods on general advertisement generation tasks. The [PITH_FULL_IMAGE:figures/full_fig_p018_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: The generated advertising images of different methods on personalized advertisement generation tasks. [PITH_FULL_IMAGE:figures/full_fig_p019_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: The generated advertising texts of different methods on general advertisement generation tasks. Some of the covered selling [PITH_FULL_IMAGE:figures/full_fig_p020_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: The generated advertising texts of different methods on personalized advertisement generation tasks. Some of the covered [PITH_FULL_IMAGE:figures/full_fig_p020_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: The generated advertising images and texts of our method on general advertisement generation tasks. Some of the covered [PITH_FULL_IMAGE:figures/full_fig_p021_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: The generated advertising images and texts of our method on personalized advertisement generation tasks. Some of the covered [PITH_FULL_IMAGE:figures/full_fig_p022_19.png] view at source ↗
read the original abstract

Generating realistic and user-preferred advertisements is a key challenge in e-commerce. Existing approaches utilize multiple independent models driven by click-through-rate (CTR) to controllably create attractive image or text advertisements. However, their pipelines lack cross-modal perception and rely on CTR that only reflects average preferences. Therefore, we explore jointly generating personalized image-text advertisements from historical click behaviors. We first design a Unified Advertisement Generative model (Uni-AdGen) that employs a single autoregressive framework to produce both advertising images and texts. By incorporating a foreground perception module and instruction tuning, Uni-AdGen enhances the realism of the generated content. To further personalize advertisements, we equip Uni-AdGen with a coarse-to-fine preference understanding module that effectively captures user interests from noisy multimodal historical behaviors to drive personalized generation. Additionally, we construct the first large-scale Personalized Advertising image-text dataset (PAd1M) and introduce a Product Background Similarity (PBS) metric to facilitate training and evaluation. Extensive experiments show that our method outperforms baselines in general and personalized advertisement generation. Our project is available at https://github.com/JD-GenX/Uni-AdGen.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 2 minor

Summary. The paper introduces Uni-AdGen, a single autoregressive model for jointly generating personalized advertising images and texts from historical click behaviors. It incorporates a foreground perception module with instruction tuning for realism and a coarse-to-fine preference understanding module to extract user interests from noisy multimodal data. The authors also release the PAd1M dataset and propose the Product Background Similarity (PBS) metric, claiming that extensive experiments demonstrate outperformance over baselines in both general and personalized ad generation tasks.

Significance. If the results hold, the work offers a practical advance in unified multimodal generation for e-commerce advertising by moving beyond separate CTR-driven pipelines to a single model conditioned on individual user history. The release of a large-scale personalized ad dataset would be a useful community resource, though its impact depends on whether the proposed PBS metric and preference module demonstrably improve perceived ad quality beyond existing metrics.

major comments (3)
  1. [§4.2] §4.2: PBS is introduced as the primary evaluation metric for background similarity between generated and reference product images, yet the manuscript provides no ablation, correlation analysis, or validation against human judgments, FID, CLIPScore, or CTR. Without evidence that PBS aligns with ad quality or user preference, the central claim of outperformance over baselines rests on an unverified metric and cannot be assessed.
  2. [Experiments] Experiments section: The abstract and main claims assert quantitative outperformance, but no specific metrics, baselines, dataset statistics, error bars, or ablation results are supplied in the provided text. This prevents verification of the reported gains and leaves the soundness of the empirical evaluation unclear.
  3. [Coarse-to-fine preference understanding module] Coarse-to-fine preference module: The assumption that historical click behaviors yield reliable personalized signals is load-bearing for the personalization contribution, but no analysis of noise robustness, bias introduction, or comparison to simpler conditioning methods is presented.
minor comments (2)
  1. [§3] The notation for the unified autoregressive framework and module interfaces could be clarified with an explicit diagram or pseudocode to aid reproducibility.
  2. [Dataset] Dataset construction details for PAd1M (e.g., filtering criteria, user count, click distribution) are referenced but not fully tabulated, which would help readers assess its scale and diversity.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment point-by-point below and will incorporate revisions to strengthen the empirical validation and analysis in the next version of the manuscript.

read point-by-point responses

  1. Referee: [§4.2] §4.2: PBS is introduced as the primary evaluation metric for background similarity between generated and reference product images, yet the manuscript provides no ablation, correlation analysis, or validation against human judgments, FID, CLIPScore, or CTR. Without evidence that PBS aligns with ad quality or user preference, the central claim of outperformance over baselines rests on an unverified metric and cannot be assessed.

    Authors: We agree that additional validation for PBS is necessary. PBS was introduced to specifically quantify background similarity for product-centric ads, complementing general metrics like FID. In the revised manuscript, we will add an ablation study removing PBS from training/evaluation, Pearson/Spearman correlations with human preference ratings on ad quality, and direct comparisons against FID, CLIPScore, and CTR on the PAd1M test set to demonstrate alignment with user-perceived ad effectiveness. revision: yes

  2. Referee: Experiments section: The abstract and main claims assert quantitative outperformance, but no specific metrics, baselines, dataset statistics, error bars, or ablation results are supplied in the provided text. This prevents verification of the reported gains and leaves the soundness of the empirical evaluation unclear.

    Authors: The full manuscript contains these details (PBS, FID, CLIPScore, human study scores; baselines including separate CTR-driven image/text models and multimodal generators; PAd1M statistics with 1M image-text pairs; error bars from 3 runs; ablations on foreground perception, instruction tuning, and coarse-to-fine modules). We will reorganize the Experiments section for clearer presentation with additional tables summarizing all numbers and statistical significance tests in the revision. revision: partial

  3. Referee: Coarse-to-fine preference module: The assumption that historical click behaviors yield reliable personalized signals is load-bearing for the personalization contribution, but no analysis of noise robustness, bias introduction, or comparison to simpler conditioning methods is presented.

    Authors: We will add this analysis in the revision. New experiments will include: (1) controlled noise injection into click histories (random clicks, missing modalities) and resulting performance degradation curves; (2) bias checks via demographic subgroup analysis on PAd1M; (3) direct comparisons to simpler baselines such as mean-pooled history embeddings or single-stage conditioning. These results will quantify the module's robustness and incremental benefit. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper describes an autoregressive model trained end-to-end on historical click data and a newly constructed PAd1M dataset, with evaluation using both standard metrics and the introduced PBS background-similarity score. No mathematical derivations, predictions, or first-principles results are presented that reduce by construction to fitted parameters or self-referential definitions. The central claims rest on empirical outperformance on held-out data rather than any self-definitional loop, fitted-input-as-prediction, or load-bearing self-citation chain. The PBS metric is an auxiliary evaluation tool whose correlation with human preference is not proven, but this is a validity concern rather than circularity in the derivation.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 2 invented entities

The central claim rests on standard transformer assumptions plus new modules whose effectiveness is asserted via experiments. No explicit free parameters beyond typical training hyperparameters are named.

free parameters (1)
  • model hyperparameters
    Standard values such as learning rate and batch size used in training the autoregressive model.
axioms (1)
  • domain assumption Autoregressive models can jointly model image and text distributions when conditioned on user history
    Invoked by the unified framework design.
invented entities (2)
  • foreground perception module no independent evidence
    purpose: Improve realism by focusing on product foregrounds
    New component added to the generative model
  • coarse-to-fine preference understanding module no independent evidence
    purpose: Extract user interests from noisy multimodal click data
    New component for personalization

pith-pipeline@v0.9.0 · 5561 in / 1260 out tokens · 43576 ms · 2026-05-13T07:31:24.760936+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

88 extracted references · 88 canonical work pages · 11 internal anchors

  1. [1]

    GPT-4 Technical Report

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ah- mad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. Gpt-4 technical report.arXiv preprint arXiv:2303.08774,

    work page internal anchor Pith review Pith/arXiv arXiv

  2. [2]

    Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, et al. Qwen2. 5-vl technical report.arXiv preprint arXiv:2502.13923, 2025. 1, 4, 7

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  3. [3]

    Improving image generation with better captions.Computer Science

    James Betker, Gabriel Goh, Li Jing, Tim Brooks, Jianfeng Wang, Linjie Li, Long Ouyang, Juntang Zhuang, Joyce Lee, Yufei Guo, et al. Improving image generation with better captions.Computer Science. https://cdn. openai. com/papers/dall-e-3. pdf, 2(3):8, 2023. 1, 2

    work page 2023

  4. [4]

    Relactrl: Relevance-guided efficient control for diffusion transformers.arXiv preprint arXiv:2502.14377,

    Ke Cao, Jing Wang, Ao Ma, Jiasong Feng, Zhanjie Zhang, Xuanhua He, Shanyuan Liu, Bo Cheng, Dawei Leng, Yuhui Yin, et al. Relactrl: Relevance-guided efficient control for diffusion transformers.arXiv preprint arXiv:2502.14377,

    work page arXiv

  5. [5]

    Prod- uct2img: Prompt-free e-commerce product background gen- eration with diffusion model and self-improved lmm

    Tingfeng Cao, Junsheng Kong, Xue Zhao, Wenqing Yao, Junwei Ding, Jinhui Zhu, and Jiandong Zhang. Prod- uct2img: Prompt-free e-commerce product background gen- eration with diffusion model and self-improved lmm. InPro- ceedings of the 32nd ACM International Conference on Mul- timedia, pages 10774–10783, 2024. 1, 2

    work page 2024

  6. [6]

    Maxmin-rlhf: Towards equitable alignment of large language models with diverse human preferences

    Souradip Chakraborty, Jiahao Qiu, Hui Yuan, Alec Koppel, Furong Huang, Dinesh Manocha, Amrit Bedi, and Mengdi Wang. Maxmin-rlhf: Towards equitable alignment of large language models with diverse human preferences. InICML 2024 Workshop on Models of Human Feedback for AI Align- ment, 2024. 4

    work page 2024

  7. [7]

    Enhancing prompt following with visual control through training-free mask-guided diffu- sion.arXiv preprint arXiv:2404.14768, 2024

    Hongyu Chen, Yiqi Gao, Min Zhou, Peng Wang, Xubin Li, Tiezheng Ge, and Bo Zheng. Enhancing prompt following with visual control through training-free mask-guided diffu- sion.arXiv preprint arXiv:2404.14768, 2024. 2

    work page arXiv 2024

  8. [8]

    T-stars- poster: A framework for product-centric advertising image design.arXiv preprint arXiv:2501.14316, 2025

    Hongyu Chen, Min Zhou, Jing Jiang, Jiale Chen, Yang Lu, Zihang Lin, Bo Xiao, Tiezheng Ge, and Bo Zheng. T-stars- poster: A framework for product-centric advertising image design.arXiv preprint arXiv:2501.14316, 2025. 1, 2

    work page arXiv 2025

  9. [9]

    Hllm-creator: Hierarchical llm- based personalized creative generation.arXiv preprint arXiv:2508.18118, 2025

    Junyi Chen, Lu Chi, Siliang Xu, Shiwei Ran, Bingyue Peng, and Zehuan Yuan. Hllm-creator: Hierarchical llm- based personalized creative generation.arXiv preprint arXiv:2508.18118, 2025. 2, 3

    work page arXiv 2025

  10. [10]

    Enhancing uncertainty modeling with semantic graph for hallucination detection

    Kedi Chen, Qin Chen, Jie Zhou, Xinqi Tao, Bowen Ding, Jingwen Xie, Mingchen Xie, Peilong Li, and Zheng Feng. Enhancing uncertainty modeling with semantic graph for hallucination detection. InProceedings of the AAAI Con- ference on Artificial Intelligence, pages 23586–23594, 2025. 2

    work page 2025

  11. [11]

    Pog: personalized outfit generation for fashion recommendation at alibaba ifashion

    Wen Chen, Pipei Huang, Jiaming Xu, Xin Guo, Cheng Guo, Fei Sun, Chao Li, Andreas Pfadler, Huan Zhao, and Bin- qiang Zhao. Pog: personalized outfit generation for fashion recommendation at alibaba ifashion. InProceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining, pages 2662–2670, 2019. 4

    work page 2019

  12. [12]

    An empirical study of training self-supervised vision transformers

    Xinlei Chen, Saining Xie, and Kaiming He. An empirical study of training self-supervised vision transformers. InPro- ceedings of the IEEE/CVF international conference on com- puter vision, pages 9640–9649, 2021. 4

    work page 2021

  13. [13]

    Ctr-driven advertising image generation with multimodal large language models

    Xingye Chen, Wei Feng, Zhenbang Du, Weizhen Wang, Yanyin Chen, Haohan Wang, Linkai Liu, Yaoyu Li, Jinyuan Zhao, Yu Li, et al. Ctr-driven advertising image generation with multimodal large language models. InProceedings of the ACM on Web Conference 2025, pages 2262–2275, 2025. 1, 2, 4

    work page 2025

  14. [14]

    Janus-Pro: Unified Multimodal Understanding and Generation with Data and Model Scaling

    Xiaokang Chen, Zhiyu Wu, Xingchao Liu, Zizheng Pan, Wen Liu, Zhenda Xie, Xingkai Yu, and Chong Ruan. Janus- pro: Unified multimodal understanding and generation with data and model scaling.arXiv preprint arXiv:2501.17811,

    work page internal anchor Pith review Pith/arXiv arXiv

  15. [15]

    Context-aware autoregressive mod- els for multi-conditional image generation.arXiv preprint arXiv:2505.12274, 2025

    Yixiao Chen, Zhiyuan Ma, Guoli Jia, Che Jiang, Jianjun Li, and Bowen Zhou. Context-aware autoregressive mod- els for multi-conditional image generation.arXiv preprint arXiv:2505.12274, 2025. 3

    work page arXiv 2025

  16. [16]

    Ctr-driven ad text generation via online feedback preference optimization

    Yanda Chen, Zihui Ren, Qixiang Gao, Jiale Chen, Si Chen, Xubin Li, Tiezheng Ge, and Bo Zheng. Ctr-driven ad text generation via online feedback preference optimization. arXiv preprint arXiv:2507.20227, 2025. 1, 2, 3

    work page arXiv 2025

  17. [17]

    Reproducible scal- ing laws for contrastive language-image learning

    Mehdi Cherti, Romain Beaumont, Ross Wightman, Mitchell Wortsman, Gabriel Ilharco, Cade Gordon, Christoph Schuh- mann, Ludwig Schmidt, and Jenia Jitsev. Reproducible scal- ing laws for contrastive language-image learning. InPro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 2818–2829, 2023. 4

    work page 2023

  18. [18]

    Towards reliable advertising image generation us- ing human feedback

    Zhenbang Du, Wei Feng, Haohan Wang, Yaoyu Li, Jingsen Wang, Jian Li, Zheng Zhang, Jingjing Lv, Xin Zhu, Junsheng Jin, et al. Towards reliable advertising image generation us- ing human feedback. InEuropean Conference on Computer Vision, pages 399–415. Springer, 2024. 1, 4, 7, 3

    work page 2024

  19. [19]

    Scaling recti- fied flow transformers for high-resolution image synthesis

    Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas M ¨uller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, et al. Scaling recti- fied flow transformers for high-resolution image synthesis. InForty-first international conference on machine learning,

    work page

  20. [20]

    Autopp: Towards automated product poster generation and optimization.arXiv preprint arXiv:2512.21921, 2025

    Jiahao Fan, Yuxin Qin, Wei Feng, Yanyin Chen, Yaoyu Li, Ao Ma, Yixiu Li, Li Zhuang, Haoyi Bian, Zheng Zhang, et al. Autopp: Towards automated product poster generation and optimization.arXiv preprint arXiv:2512.21921, 2025. 2

    work page arXiv 2025

  21. [21]

    Postermaker: Towards 9 high-quality product poster generation with accurate text rendering

    Yifan Gao, Zihang Lin, Chuanbin Liu, Min Zhou, Tiezheng Ge, Bo Zheng, and Hongtao Xie. Postermaker: Towards 9 high-quality product poster generation with accurate text rendering. InProceedings of the Computer Vision and Pat- tern Recognition Conference, pages 8083–8093, 2025. 1, 2, 4, 7, 3

    work page 2025

  22. [22]

    DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning

    Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, et al. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning.arXiv preprint arXiv:2501.12948, 2025. 1, 2, 3, 7, 8, 4

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  23. [23]

    Controlthinker: Unveiling latent semantics for controllable image generation through visual reasoning.arXiv preprint arXiv:2506.03596, 2025

    Feng Han, Yang Jiao, Shaoxiang Chen, Junhao Xu, Jingjing Chen, and Yu-Gang Jiang. Controlthinker: Unveiling latent semantics for controllable image generation through visual reasoning.arXiv preprint arXiv:2506.03596, 2025. 3

    work page arXiv 2025

  24. [24]

    The movielens datasets: History and context.Acm transactions on interac- tive intelligent systems (tiis), 5(4):1–19, 2015

    F Maxwell Harper and Joseph A Konstan. The movielens datasets: History and context.Acm transactions on interac- tive intelligent systems (tiis), 5(4):1–19, 2015. 4

    work page 2015

  25. [25]

    Lora: Low-rank adaptation of large language models.ICLR, 1(2):3, 2022

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen- Zhu, Yuanzhi Li, Shean Wang, Lu Wang, Weizhu Chen, et al. Lora: Low-rank adaptation of large language models.ICLR, 1(2):3, 2022. 7

    work page 2022

  26. [26]

    Improving generative ad text on facebook using reinforcement learning.arXiv preprint arXiv:2507.21983, 2025

    Daniel R Jiang, Alex Nikulkov, Yu-Chia Chen, Yang Bai, and Zheqing Zhu. Improving generative ad text on facebook using reinforcement learning.arXiv preprint arXiv:2507.21983, 2025. 2, 3

    work page arXiv 2025

  27. [27]

    Towards improving coherence and diversity of slogan generation.Natural Language Engineering, 29(2):254–286,

    Yiping Jin, Akshay Bhatia, Dittaya Wanvarie, and Phu TV Le. Towards improving coherence and diversity of slogan generation.Natural Language Engineering, 29(2):254–286,

    work page

  28. [28]

    Cobart: controlled, optimized, bidirectional and auto- regressive transformer for ad headline generation

    Yashal Shakti Kanungo, Gyanendra Das, Pooja A, and Sumit Negi. Cobart: controlled, optimized, bidirectional and auto- regressive transformer for ad headline generation. InPro- ceedings of the 28th ACM SIGKDD conference on knowl- edge discovery and data mining, pages 3127–3136, 2022. 2

    work page 2022

  29. [29]

    Pick-a-pic: An open dataset of user preferences for text-to-image generation.Ad- vances in neural information processing systems, 36:36652– 36663, 2023

    Yuval Kirstain, Adam Polyak, Uriel Singer, Shahbuland Ma- tiana, Joe Penna, and Omer Levy. Pick-a-pic: An open dataset of user preferences for text-to-image generation.Ad- vances in neural information processing systems, 36:36652– 36663, 2023. 7

    work page 2023

  30. [30]

    Flux.https://github.com/ black-forest-labs/flux, 2024

    Black Forest Labs. Flux.https://github.com/ black-forest-labs/flux, 2024. 1, 2, 4, 7, 3

    work page 2024

  31. [31]

    Black Forest Labs, Stephen Batifol, Andreas Blattmann, Frederic Boesel, Saksham Consul, Cyril Diagne, Tim Dock- horn, Jack English, Zion English, Patrick Esser, et al. Flux. 1 kontext: Flow matching for in-context image generation and editing in latent space.arXiv preprint arXiv:2506.15742,

    work page internal anchor Pith review Pith/arXiv arXiv

  32. [32]

    Relation-aware diffusion model for controllable poster layout generation

    Fengheng Li, An Liu, Wei Feng, Honghe Zhu, Yaoyu Li, Zheng Zhang, Jingjing Lv, Xin Zhu, Junjie Shen, Zhangang Lin, et al. Relation-aware diffusion model for controllable poster layout generation. InProceedings of the 32nd ACM International Conference on Information and Knowledge Management, pages 1249–1258, 2023. 2

    work page 2023

  33. [33]

    Culg: Commer- cial universal language generation

    Haonan Li, Yameng Huang, Yeyun Gong, Jian Jiao, Ruofei Zhang, Timothy Baldwin, and Nan Duan. Culg: Commer- cial universal language generation. InProceedings of the 2022 Conference of the North American Chapter of the As- sociation for Computational Linguistics: Human Language Technologies: Industry Track, pages 112–120, 2022. 1, 2

    work page 2022

  34. [34]

    Controlvar: Exploring con- trollable visual autoregressive modeling.arXiv preprint arXiv:2406.09750, 2024

    Xiang Li, Kai Qiu, Hao Chen, Jason Kuen, Zhe Lin, Rita Singh, and Bhiksha Raj. Controlvar: Exploring con- trollable visual autoregressive modeling.arXiv preprint arXiv:2406.09750, 2024. 3

    work page arXiv 2024

  35. [35]

    Planning and rendering: Towards prod- uct poster generation with diffusion models.arXiv preprint arXiv:2312.08822, 2023

    Zhaochen Li, Fengheng Li, Wei Feng, Honghe Zhu, Yaoyu Li, Zheng Zhang, Jingjing Lv, Junjie Shen, Zhangang Lin, Jingping Shao, et al. Planning and rendering: Towards prod- uct poster generation with diffusion models.arXiv preprint arXiv:2312.08822, 2023. 1, 2

    work page arXiv 2023

  36. [36]

    Controlar: Controllable image generation with autoregressive models.arXiv preprint arXiv:2410.02705, 2024

    Zongming Li, Tianheng Cheng, Shoufa Chen, Peize Sun, Haocheng Shen, Longjin Ran, Xiaoxin Chen, Wenyu Liu, and Xinggang Wang. Controlar: Controllable image generation with autoregressive models.arXiv preprint arXiv:2410.02705, 2024. 1, 3

    work page arXiv 2024

  37. [37]

    Rouge: A package for automatic evaluation of summaries

    Chin-Yew Lin. Rouge: A package for automatic evaluation of summaries. InText summarization branches out, pages 74–81, 2004. 4, 7

    work page 2004

  38. [38]

    Autoposter: A highly automatic and content-aware design system for ad- vertising poster generation

    Jinpeng Lin, Min Zhou, Ye Ma, Yifan Gao, Chenxi Fei, Yangjian Chen, Zhang Yu, and Tiezheng Ge. Autoposter: A highly automatic and content-aware design system for ad- vertising poster generation. InProceedings of the 31st ACM International Conference on Multimedia, pages 1250–1260,

    work page

  39. [39]

    Ragar: Retrieval augment person- alized image generation guided by recommendation.arXiv preprint arXiv:2505.01657, 2025

    Run Ling, Wenji Wang, Yuting Liu, Guibing Guo, Haowei Liu, Jian Lu, Quanwei Zhang, Yexing Xu, Shuo Lu, Yun Wang, et al. Ragar: retrieval augmented personalized im- age generation guided by recommendation.arXiv preprint arXiv:2505.01657, 2025. 4

    work page arXiv 2025

  40. [40]

    Ser30k: A large-scale dataset for sticker emotion recognition

    Shengzhe Liu, Xin Zhang, and Jufeng Yang. Ser30k: A large-scale dataset for sticker emotion recognition. InPro- ceedings of the 30th ACM International Conference on Mul- timedia, pages 33–41, 2022. 4

    work page 2022

  41. [41]

    Uni-layout: Integrating human feedback in unified layout generation and evaluation

    Shuo Lu, Yanyin Chen, Wei Feng, Jiahao Fan, Fengheng Li, Zheng Zhang, Jingjing Lv, Junjie Shen, Ching Law, and Jian Liang. Uni-layout: Integrating human feedback in unified layout generation and evaluation. InProceedings of the 33rd ACM International Conference on Multimedia, pages 7709– 7718, 2025. 2

    work page 2025

  42. [42]

    One size, many fits: Aligning diverse group-wise click preferences in large-scale advertising image generation

    Shuo Lu, Haohan Wang, Wei Feng, Weizhen Wang, Shen Zhang, Yaoyu Li, Ao Ma, Zheng Zhang, Jingjing Lv, Junjie Shen, et al. One size, many fits: Aligning diverse group-wise click preferences in large-scale advertising image generation. arXiv preprint arXiv:2602.02033, 2026. 2

    work page arXiv 2026

  43. [43]

    Lay2story: extending diffu- sion transformers for layout-togglable story generation

    Ao Ma, Jiasong Feng, Ke Cao, Jing Wang, Yun Wang, Quan- wei Zhang, and Zhanjie Zhang. Lay2story: extending diffu- sion transformers for layout-togglable story generation. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 16102–16111, 2025. 4

    work page 2025

  44. [44]

    Adparaphrase: Para- phrase dataset for analyzing linguistic features toward gen- erating attractive ad texts.arXiv preprint arXiv:2502.04674,

    Soichiro Murakami, Peinan Zhang, Hidetaka Kamigaito, Hi- roya Takamura, and Manabu Okumura. Adparaphrase: Para- phrase dataset for analyzing linguistic features toward gen- erating attractive ad texts.arXiv preprint arXiv:2502.04674,

    work page arXiv

  45. [45]

    DINOv2: Learning Robust Visual Features without Supervision

    Maxime Oquab, Timoth ´ee Darcet, Th ´eo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193, 2023. 6 10

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  46. [46]

    Bleu: a method for automatic evaluation of machine translation

    Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. Bleu: a method for automatic evaluation of machine translation. InProceedings of the 40th annual meeting of the Association for Computational Linguistics, pages 311–318,

    work page

  47. [47]

    SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis

    Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas M ¨uller, Joe Penna, and Robin Rombach. Sdxl: Improving latent diffusion mod- els for high-resolution image synthesis.arXiv preprint arXiv:2307.01952, 2023. 2, 4

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  48. [48]

    Qwen2.5 technical report, 2025

    Qwen, :, An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Haoran Wei, Huan Lin, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxi Yang, Jingren Zhou, Jun- yang Lin, Kai Dang, Keming Lu, Keqin Bao, Kexin Yang, Le Yu, Mei Li, Mingfeng Xue, Pei Zhang, Qin Zhu, Rui Men, Runji Lin, Tianhao L...

    work page 2025

  49. [49]

    Language models are unsu- pervised multitask learners.OpenAI blog, 1(8):9, 2019

    Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, Ilya Sutskever, et al. Language models are unsu- pervised multitask learners.OpenAI blog, 1(8):9, 2019. 1

    work page 2019

  50. [50]

    Hierarchical Text-Conditional Image Generation with CLIP Latents

    Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen. Hierarchical text-conditional image gener- ation with clip latents.arXiv preprint arXiv:2204.06125, 1 (2):3, 2022. 1, 2

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  51. [51]

    Group robust preference optimization in reward- free rlhf.Advances in Neural Information Processing Sys- tems, 37:37100–37137, 2024

    Shyam Sundhar Ramesh, Yifan Hu, Iason Chaimalas, Viraj Mehta, Pier Giuseppe Sessa, Haitham Bou Ammar, and Ilija Bogunovic. Group robust preference optimization in reward- free rlhf.Advances in Neural Information Processing Sys- tems, 37:37100–37137, 2024. 4

    work page 2024

  52. [52]

    Grounded sam: Assembling open-world models for diverse visual tasks,

    Tianhe Ren, Shilong Liu, Ailing Zeng, Jing Lin, Kun- chang Li, He Cao, Jiayu Chen, Xinyu Huang, Yukang Chen, Feng Yan, Zhaoyang Zeng, Hao Zhang, Feng Li, Jie Yang, Hongyang Li, Qing Jiang, and Lei Zhang. Grounded sam: Assembling open-world models for diverse visual tasks,

    work page

  53. [53]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer. High-resolution image synthesis with latent diffusion models. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022. 2

    work page 2022

  54. [54]

    Clip+mlp aesthetic score pre- dictor

    Christoph Schuhmann. Clip+mlp aesthetic score pre- dictor. https://github.com/christophschuhmann/improved- aesthetic-predictor., 2022. 7

    work page 2022

  55. [55]

    Controllable and diverse text generation in e-commerce

    Huajie Shao, Jun Wang, Haohong Lin, Xuezhou Zhang, As- ton Zhang, Heng Ji, and Tarek Abdelzaher. Controllable and diverse text generation in e-commerce. InProceedings of the Web Conference 2021, pages 2392–2401, 2021. 1, 2

    work page 2021

  56. [56]

    Autoprompt: Eliciting knowledge from language models wit h automatically generated prompts,

    Taylor Shin, Yasaman Razeghi, Robert L Logan IV , Eric Wallace, and Sameer Singh. Autoprompt: Eliciting knowl- edge from language models with automatically generated prompts.arXiv preprint arXiv:2010.15980, 2020. 1

    work page arXiv 2010

  57. [57]

    DINOv3

    Oriane Sim ´eoni, Huy V V o, Maximilian Seitzer, Federico Baldassarre, Maxime Oquab, Cijo Jose, Vasil Khalidov, Marc Szafraniec, Seungeun Yi, Micha ¨el Ramamonjisoa, et al. Dinov3.arXiv preprint arXiv:2508.10104, 2025. 4

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  58. [58]

    Autoregressive Model Beats Diffusion: Llama for Scalable Image Generation

    Peize Sun, Yi Jiang, Shoufa Chen, Shilong Zhang, Bingyue Peng, Ping Luo, and Zehuan Yuan. Autoregressive model beats diffusion: Llama for scalable image generation.arXiv preprint arXiv:2406.06525, 2024. 5

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  59. [59]

    Minimal impact controlnet: Advancing multi-controlnet integration.arXiv preprint arXiv:2506.01672, 2025

    Shikun Sun, Min Zhou, Zixuan Wang, Xubin Li, Tiezheng Ge, Zijie Ye, Xiaoyu Qin, Junliang Xing, Bo Zheng, and Jia Jia. Minimal impact controlnet: Advancing multi-controlnet integration.arXiv preprint arXiv:2506.01672, 2025. 2

    work page arXiv 2025

  60. [60]

    Attention is all you need.Advances in neural information processing systems, 30, 2017

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszko- reit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need.Advances in neural information processing systems, 30, 2017. 6

    work page 2017

  61. [61]

    Beyond quality: Unlocking diver- sity in ad headline generation with large language models

    Chang Wang, Siyu Yan, Depeng Yuan, Yuqi Chen, Yan- hua Huang, Yuanhang Zheng, Shuhao Li, Yinqi Zhang, Kedi Chen, Mingrui Zhu, et al. Beyond quality: Unlocking diver- sity in ad headline generation with large language models. arXiv preprint arXiv:2508.18739, 2025. 1, 2, 3

    work page arXiv 2025

  62. [62]

    Gen- erate e-commerce product background by integrating cate- gory commonality and personalized style

    Haohan Wang, Wei Feng, Yaoyu Li, Zheng Zhang, Jingjing Lv, Junjie Shen, Zhangang Lin, and Jingping Shao. Gen- erate e-commerce product background by integrating cate- gory commonality and personalized style. InICASSP 2025- 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 1–5. IEEE, 2025. 2

    work page 2025

  63. [63]

    Wisa: World simulator assistant for physics-aware text-to-video generation.arXiv preprint arXiv:2503.08153, 2025

    Jing Wang, Ao Ma, Ke Cao, Jun Zheng, Zhanjie Zhang, Jiasong Feng, Shanyuan Liu, Yuhang Ma, Bo Cheng, Dawei Leng, et al. Wisa: World simulator assistant for physics-aware text-to-video generation.arXiv preprint arXiv:2503.08153, 2025. 4

    work page arXiv 2025

  64. [64]

    Imagen editor and editbench: Advancing and evaluating text-guided im- age inpainting

    Su Wang, Chitwan Saharia, Ceslee Montgomery, Jordi Pont- Tuset, Shai Noy, Stefano Pellegrini, Yasumasa Onoe, Sarah Laszlo, David J Fleet, Radu Soricut, et al. Imagen editor and editbench: Advancing and evaluating text-guided im- age inpainting. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 18359– 18369, 2023. 1, 2

    work page 2023

  65. [65]

    Adstereo: Efficient stereo matching with adaptive downsampling and disparity alignment.IEEE Transactions on Image Processing, 2025

    Yun Wang, Kunhong Li, Longguang Wang, Junjie Hu, Dapeng Oliver Wu, and Yulan Guo. Adstereo: Efficient stereo matching with adaptive downsampling and disparity alignment.IEEE Transactions on Image Processing, 2025. 4

    work page 2025

  66. [66]

    Dualnet: Ro- bust self-supervised stereo matching with pseudo-label su- pervision

    Yun Wang, Jiahao Zheng, Chenghao Zhang, Zhanjie Zhang, Kunhong Li, Yongjian Zhang, and Junjie Hu. Dualnet: Ro- bust self-supervised stereo matching with pseudo-label su- pervision. InProceedings of the AAAI Conference on Artifi- cial Intelligence, pages 8178–8186, 2025. 4

    work page 2025

  67. [67]

    Creater: Ctr-driven advertising text genera- tion with controlled pre-training and contrastive fine-tuning

    Penghui Wei, Xuanhua Yang, Shaoguo Liu, Liang Wang, and Bo Zheng. Creater: Ctr-driven advertising text genera- tion with controlled pre-training and contrastive fine-tuning. arXiv preprint arXiv:2205.08943, 2022. 1, 2

    work page arXiv 2022

  68. [68]

    Janus: Decoupling visual encod- ing for unified multimodal understanding and generation

    Chengyue Wu, Xiaokang Chen, Zhiyu Wu, Yiyang Ma, Xingchao Liu, Zizheng Pan, Wen Liu, Zhenda Xie, Xingkai Yu, Chong Ruan, et al. Janus: Decoupling visual encod- ing for unified multimodal understanding and generation. In Proceedings of the Computer Vision and Pattern Recognition Conference, pages 12966–12977, 2025. 1, 5

    work page 2025

  69. [69]

    Lumina-mgpt 2.0: Stand-alone autoregressive image modeling.arXiv preprint arXiv:2507.17801, 2025

    Yi Xin, Juncheng Yan, Qi Qin, Zhen Li, Dongyang Liu, Shicheng Li, Victor Shea-Jay Huang, Yupeng Zhou, Ren- 11 rui Zhang, Le Zhuo, et al. Lumina-mgpt 2.0: Stand- alone autoregressive image modeling.arXiv preprint arXiv:2507.17801, 2025. 3

    work page arXiv 2025

  70. [70]

    Imagere- ward: Learning and evaluating human preferences for text- to-image generation.Advances in Neural Information Pro- cessing Systems, 36:15903–15935, 2023

    Jiazheng Xu, Xiao Liu, Yuchen Wu, Yuxuan Tong, Qinkai Li, Ming Ding, Jie Tang, and Yuxiao Dong. Imagere- ward: Learning and evaluating human preferences for text- to-image generation.Advances in Neural Information Pro- cessing Systems, 36:15903–15935, 2023. 7

    work page 2023

  71. [71]

    Scalar: Scale-wise controllable visual autoregressive learning.arXiv preprint arXiv:2507.19946, 2025

    Ryan Xu, Dongyang Jin, Yancheng Bai, Rui Lan, Xu Duan, Lei Sun, and Xiangxiang Chu. Scalar: Scale-wise controllable visual autoregressive learning.arXiv preprint arXiv:2507.19946, 2025. 3

    work page arXiv 2025

  72. [72]

    Personalized image generation with large multimodal models

    Yiyan Xu, Wenjie Wang, Yang Zhang, Biao Tang, Peng Yan, Fuli Feng, and Xiangnan He. Personalized image generation with large multimodal models. InProceedings of the ACM on Web Conference 2025, pages 264–274, 2025. 8, 4

    work page 2025

  73. [73]

    Qwen3 Technical Report

    An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, et al. Qwen3 technical report.arXiv preprint arXiv:2505.09388, 2025. 1, 2, 3, 7, 8, 4

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  74. [74]

    A new creative generation pipeline for click- through rate with stable diffusion model

    Hao Yang, Jianxin Yuan, Shuai Yang, Linhe Xu, Shuo Yuan, and Yifan Zeng. A new creative generation pipeline for click- through rate with stable diffusion model. InCompanion Pro- ceedings of the ACM Web Conference 2024, pages 180–189,

    work page 2024

  75. [75]

    Car: Controllable autoregressive modeling for visual generation

    Ziyu Yao, Jialin Li, Yifeng Zhou, Yong Liu, Xi Jiang, Chengjie Wang, Feng Zheng, Yuexian Zou, and Lei Li. Car: Controllable autoregressive modeling for visual generation. arXiv preprint arXiv:2410.04671, 2024. 3

    work page arXiv 2024

  76. [76]

    Attribute alignment: Controlling text generation from pre-trained language mod- els.arXiv preprint arXiv:2103.11070, 2021

    Dian Yu, Zhou Yu, and Kenji Sagae. Attribute alignment: Controlling text generation from pre-trained language mod- els.arXiv preprint arXiv:2103.11070, 2021. 1

    work page arXiv 2021

  77. [77]

    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 Proceedings of the IEEE/CVF international conference on computer vision, pages 3836–3847, 2023. 1, 2, 3

    work page 2023

  78. [78]

    U- stydit: Ultra-high quality artistic style transfer using diffu- sion transformers.arXiv preprint arXiv:2503.08157, 2025

    Zhanjie Zhang, Ao Ma, Ke Cao, Jing Wang, Shanyuan Liu, Yuhang Ma, Bo Cheng, Dawei Leng, and Yuhui Yin. U- stydit: Ultra-high quality artistic style transfer using diffu- sion transformers.arXiv preprint arXiv:2503.08157, 2025. 4

    work page arXiv 2025

  79. [79]

    Dreampainter: Image background inpainting for e-commerce scenarios.arXiv preprint arXiv:2508.02155,

    Sijie Zhao, Jing Cheng, Yaoyao Wu, Hao Xu, and Shao- hui Jiao. Dreampainter: Image background inpainting for e-commerce scenarios.arXiv preprint arXiv:2508.02155,

    work page arXiv

  80. [80]

    2 12 Design Your Ad: Personalized Advertising Image and Text Generation with Unified Autoregressive Models Appendix This appendix details the design for human evaluation met- rics in Sec. 7. Section 9 provides comprehensive visual com- parisons with baseline methods under both personalized and non- personalized settings. To justify our model configuration...

    work page

Showing first 80 references.