Pith. sign in

REVIEW 4 major objections 4 minor 45 references

InnoText claims that a single diffusion-transformer model can both generate and edit legible text in images across English and Chinese, outperforming task-specific systems on accuracy and visual quality.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 05:45 UTC pith:FNRFZUIM

load-bearing objection A credible unified DiT for bilingual text generation and editing, but the Chinese generation benchmark is partly self-referential and the lack of error bars makes the headline gains uncertain. the 4 major comments →

arxiv 2607.22101 v1 pith:FNRFZUIM submitted 2026-07-24 cs.CV

InnoText: A Unified Model for Visual Text Generation and Editing

Haowei Liu , Runze He , Jian Lu , Ao Ma , Run Ling , Ke Cao , Jiasong Feng , Wei Feng
show 5 more authors
Shuo Lu Yexing Xu Yun Wang Jing Wang Zhanjie Zhang
This is my paper
classification cs.CV
keywords visual text generationvisual text editingdiffusion transformer (DiT)font-size-aware modulationsmall-character augmentationbilingual English-Chinese datasetscene text renderingunified model
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

InnoText sets out to show that one Diffusion Transformer can handle both visual text generation (adding new text to an image from a prompt) and visual text editing (replacing text while keeping background and style) without losing to models specialized for only one task. The paper argues that existing U-Net pipelines mangle small or densely structured characters, especially in Chinese, while current transformer editors are stuck on editing alone. Its answer is a unified input format plus three ingredients: a Font Size-Aware Modulation that re-weights features by local font size, a Small-Character Aware Augmentation that magnifies tiny glyph regions during training, and a Task-Specific Region Weighted Loss that sharpens attention to edited areas. A new bilingual 30K-image dataset (English and Chinese) supports training and a two-language benchmark for evaluation. Reported results place this single model ahead of earlier systems on exact-text accuracy and perceptual quality in both languages.

Core claim

The paper's central claim is that generation and editing are the same underlying operation once the model is told what to preserve. The input stacks a glyph image (the desired text, rendered) with either a masked image for editing or a blank canvas for generation, and a mask channel tells the model which half to paint. A font-size map—a grayscale image of inverse character heights—is injected into each transformer block to recalibrate features, so small characters are not neglected. A small-character augmentation and a region-weighted loss keep tiny strokes legible and focus optimization on edited pixels. Fine-tuned on 30K curated bilingual images, the model reports the best exact-text accur

What carries the argument

The central device is the Font Size-Aware Modulation (FSAM): a spatial size map encoding inverse per-character height, patchified into tokens, linearly projected, and decoded by two MLPs into per-location scale and shift parameters that multiply and add into the hidden states of each Diffusion Transformer (DiT) block. This makes the network's internal gain depend on local font scale, counteracting the degradation of tiny glyphs. Two supporting mechanisms carry the same idea further: Small-Character Aware Augmentation randomly enlarges small-character regions during training (a 'foveal' zoom), and the Task-Specific Region Weighted Loss multiplies the flow-matching loss by the size map for edi

Load-bearing premise

The whole advantage rests on the assumption that the synthetic Chinese images used for training and evaluation resemble real-world Chinese text scenes closely enough; if they contain hidden rendering artifacts, the measured gains could be an artifact of the benchmark itself.

What would settle it

Hold out a set of real-world photos containing Chinese signs and labels, with varied fonts, lighting, and perspective, and run the model without any fine-tuning on that set; if OCR sentence accuracy on these real photos drops to the level of editing-only baselines, the reported edge is tied to the synthetic distribution rather than to the architecture.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If the central claim holds, one trained checkpoint can switch between generation and editing by changing only the input mask, removing the need for separate pipelines and their style drift.
  • The font-size map is language-agnostic, so the same recipe should transfer to other complex scripts such as Japanese or Korean with minimal re-curation.
  • Applying the region-weighted loss beyond text—e.g., to logo or caption edits—becomes a natural extension for other precision-critical image editing tasks.
  • The 30K dataset, if public, gives the field a common bilingual benchmark that separates text accuracy from perceptual quality.

Where Pith is reading between the lines

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

  • A test not run in the paper: evaluating on real photographs of street signs or product labels would reveal whether the synthetic training distribution is the reason the model wins, since the proposed benchmark samples the same corpus used for fine-tuning.
  • The inverse-font-height map could be reused as a general 'importance map' for any dense, small-scale visual content (charts, barcodes, fine patterns), not just text.
  • If the unified model also holds up under instruction prompts (e.g., 'change X to Y'), the same architecture could evolve into a text-editing assistant rather than a two-mode tool.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper proposes InnoText, a Diffusion-Transformer (DiT) model that unifies visual text generation and editing in a single network by building on Flux.1 Fill dev. Three components are introduced: Font Size-Aware Modulation (FSAM), Small-Character Aware Augmentation (SCAA), and a Task-Specific Region Weighted Loss. The authors also construct InnoText-30K, a bilingual English/Chinese dataset, and InnoText-Bench for evaluation. Experiments compare with AnyText, AnyText2, TextFlux, Flux-Fill, and, in the supplementary material, Flux-Text and EasyText, reporting sentence accuracy, NED, LPIPS, and human-preference results. The main claim is state-of-the-art performance on both generation and editing in English and Chinese.

Significance. The paper addresses a relevant gap: most strong DiT-based text models are single-task. If the claimed results hold, a single model that handles both generation and editing at good accuracy, with a new bilingual dataset, would be a useful contribution. The external editing benchmark (AnyText-Bench) and MARIO-Eval results provide some independent support, and the ablation study gives evidence for the contributions of FSAM and SCAA. However, the headline generation results, especially for Chinese, are evaluated on a benchmark sampled from the authors' own training corpus, and the main baseline comparisons are not matched in training data or compute. These issues currently leave the central generalization claim insufficiently supported.

major comments (4)
  1. [§3.2, Table 1] InnoText-Bench is sampled from InnoText-30K, the same corpus used to train InnoText. The Chinese subset is generated by Seedream from translated/augmented Lex-10K prompts and filtered with PPOCR and DeepSeek-VL2; no evidence is provided that this synthetic distribution matches real Chinese scene text. The reported Chinese generation advantage over AnyText2 (Sen. Acc 0.5907 vs 0.5262) may therefore be inflated by train/test overlap. The NIMA/NIQE scores in Fig. 4 do not measure text realism or glyph distribution. Please add an externally curated Chinese generation test set or a cross-dataset evaluation, and clarify how the InnoText-Bench split avoids training overlap.
  2. [§5.1, Tables 1 and 6] The main comparisons use baselines with their officially released pretrained weights; only AnyText and AnyText2 are fine-tuned on InnoText-30K in the ablation of Table 2. TextFlux, Flux-Fill, EasyText, and Flux-Text are not trained on the same data or with the same LoRA/compute budget in the headline tables. Since data composition and training volume strongly affect text rendering, the claimed superiority may reflect data/training disparities rather than a fundamental architectural advantage. The matched comparison in Table 6 (Ours 3M vs Flux-Text 3M) is more convincing; please extend this matched design to all headline baselines, including generation, or state explicitly where this is infeasible.
  3. [§4.4, Eq. (10)] The Task-Specific Region Weighted Loss is underspecified. The region weight S is not defined; the text says higher weights are assigned to masked or modified areas, but no formula, normalization, or dependence on the size map/mask is given. The flow-matching loss L_FM is also not explicitly defined. Without these definitions the method cannot be reproduced, and the ablation row 'w/o L_T_SRW' in Table 2 is not interpretable. Please provide the full loss expression, including how S is computed and normalized.
  4. [Tables 1–6 and Fig. 10] No error bars, confidence intervals, or significance tests are reported. Several headline gaps are small: e.g., English editing Sen. Acc 0.7988 vs 0.7893 for AnyText2, Ours (3M) vs Flux-Text (3M) 0.8224 vs 0.8175, and some supplementary human-preference comparisons are near chance (e.g., 51.9% vs 48.1%). Given the evaluation set sizes (1,500 images for InnoText-Bench; 100 cases in the main human study), these differences may be within sampling noise. Please report bootstrap confidence intervals or paired significance tests for the central claims.
minor comments (4)
  1. [§5.1 vs §4.3] Notation is inconsistent: §4.3 defines the small-character threshold as τ, the base resize factor as λ, and the perturbation range as ε; §5.1 says 'the size threshold λ and scaling factor ε are set to 1.5 and 0.3.' Please align the notation and specify which values correspond to τ, λ, and ε.
  2. [§5.3] The main text refers to 'Fig. 11' and 'Fig. 12' for qualitative comparisons, but those figures appear only in the supplementary material; the corresponding main-paper figures are Fig. 6 and Fig. 7. Please fix the cross-references.
  3. [Supplementary §4.2] Typo: 'Nagetive Effect' should be 'Negative Effect.'
  4. [§3.2] Please state explicitly how the 1,500-image InnoText-Bench is split from the 30K training corpus and whether any sampled images were used during training, fine-tuning, or curation filtering. This is important for interpreting all results that use InnoText-Bench.

Circularity Check

0 steps flagged

No significant derivation-level circularity; the self-referential InnoText-Bench is an evaluation-validity concern, not an equation-level circular step.

full rationale

No equation-level circularity: the contributions are architectural modules (FSAM, SCAA, task-specific loss) and an empirical evaluation; no quantity is defined in terms of the very quantity it is used to predict, and no fitted parameter is relabeled as a prediction. The self-citations (RelaCtrl, MoFu, InnoAds-Composer, etc.) appear only in related-work enumerations and are not load-bearing for the central claims. The main potential concern is that the generation benchmark, InnoText-Bench, is sampled from the same InnoText-30K corpus used for training, and the Chinese subset is Seedream-synthesized; if these 1,500 samples were not held out, the Chinese generation numbers in Table 1 would be in-distribution and could be inflated. However, the paper also reports editing on the external AnyText-Benchmark and generation on the external MARIO-Eval (Supplementary Table 4), so the central claims have independent checks. Because the benchmark issue is an evaluation-validity concern rather than a derivation-level circularity, the circularity score is 2.

Axiom & Free-Parameter Ledger

7 free parameters · 4 axioms · 0 invented entities

The central claim rests on a pretrained Flux-Fill backbone, a heuristic size map from OCR bounding boxes, and several undisclosed hyperparameters (gamma_max, k, lambda_shift, tau). The dataset construction itself depends on a generative model (Seedream) whose fidelity is assumed. These are reasonable engineering assumptions but reduce the independence of the evaluation.

free parameters (7)
  • gamma_max
    FSAM maximum amplification ratio in Eq. (6). Value not reported.
  • k
    FSAM response sharpness in Eq. (6). Value not reported.
  • lambda_shift
    Shift modulation amplitude in Eq. (7). Value not reported.
  • tau (small-character threshold)
    Threshold below which a character is considered small in SCAA (§4.3). No numeric value given; §5.1 lists lambda=1.5 and epsilon=0.3, which are different quantities.
  • lambda (SCAA base resize factor) = 1.5
    Base magnification factor in Eq. (9), set in §5.1.
  • epsilon (SCAA perturbation range) = 0.3
    Uniform perturbation range in Eq. (9), set in §5.1.
  • LoRA rank = 128
    Final rank chosen after testing 32, 64, 128, 256 (Table 5).
axioms (4)
  • domain assumption Flux-Fill pretrained model provides a strong generative prior and is adapted via LoRA
    The entire method is a fine-tune of Flux-Fill. If the backbone lacks Chinese glyph knowledge, LoRA fine-tuning may not close the gap.
  • domain assumption OCR bounding boxes used to build the size map are accurate
    Eq. (5) computes Fh from bounding boxes; if the detector (not specified) produces wrong boxes, FSAM mis-modulates features.
  • ad hoc to paper Inverse font height is a suitable proxy for 'attention needed'
    The heuristic that smaller text needs stronger modulation is plausible but not derived or validated independently.
  • standard math Flow matching loss is a valid training objective
    Standard loss for rectified flow / Flux models; used without modification.

pith-pipeline@v1.3.0-alltime-deepseek · 14181 in / 9533 out tokens · 101098 ms · 2026-08-01T05:45:15.037434+00:00 · methodology

0 comments
read the original abstract

Diffusion models have recently achieved remarkable success in high-fidelity image synthesis, yet their application to visual text generation and editing remains relatively underexplored. Unlike general image generation, visual text tasks demand precise structural regularity and legibility, which may pose additional challenges for small-scale text and non-Latin scripts such as Chinese. Existing UNet-based models often struggle to produce clear and coherent text, while DiT-based models, though more expressive, are typically limited to a single task, which may lead to redundant training pipelines, inconsistent visual styles, and reduced cross-task generalization. To address these challenges, we propose InnoText, a unified DiT-based framework capable of performing both text generation and editing within a single model. We introduce a Font Size-Aware Modulation (FSAM) module to enhance representations across font scales, a Small-Character Aware Augmentation strategy to improve fine-grained fidelity, and a Task-Specific Region Weighted Loss for adaptive optimization. To support training and evaluation, we also construct a high-quality bilingual (English-Chinese) visual text dataset covering diverse fonts, sizes, and backgrounds. Experimental results demonstrate that our method achieves superior generation accuracy and editing quality, producing visually appealing and realistic text images.

Figures

Figures reproduced from arXiv: 2607.22101 by Ao Ma, Haowei Liu, Jian Lu, Jiasong Feng, Jing Wang, Ke Cao, Run Ling, Runze He, Shuo Lu, Wei Feng, Yexing Xu, Yun Wang, Zhanjie Zhang.

Figure 1
Figure 1. Figure 1: Visual examples generated by our method for visual text generation and editing. Abstract. Diffusion models have recently achieved remarkable success in high-fidelity image synthesis, yet their application to visual text gener- ∗Equal contribution. †Corresponding author. ‡Project leader. §Conducted during internship. arXiv:2607.22101v1 [cs.CV] 24 Jul 2026 [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: (Left) Visual text generation in previous unified models [30,43]. (Middle) Visual text generation in AnyText [27] and ours. (Right) Visual text editing task [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Visual text images generated by diffusion models based on UNet and DiT, respectively. tasks including customized image generation [1, 21], visual editing [23, 40], and visual text synthesis [14, 32]. 2.2 Controllable Diffusion Transformers Traditional UNet-based generation methods, such as ControlNet [38], IP-Adapter [35], and T2I-Adapter [18], enable downstream applications by allowing the model to take r… view at source ↗
Figure 4
Figure 4. Figure 4: Datasets overview. (Left) Representative cases from multiple datasets. (Top right) Comparison of aesthetics and naturalness across different datasets. The aesthetic scores are obtained using the Neural Image Assessment (NIMA) model [24], while the naturalness scores are evaluated using the Natural Image Quality Evaluator (NIQE) metric [17]. (Bottom right) Dataset processing pipeline. vision-language recali… view at source ↗
Figure 5
Figure 5. Figure 5: Overview of InnoText. The model adopts different input paradigms and size maps for the editing and generation modes. In our framework, we incorporate a Font Size-Aware Modulation (FSAM) module that dynamically enhances the model’s attention to text regions of varying font sizes. In addition, we adopt a Small-Character Aware Augmentation strategy, which selectively enlarges small-font text samples to improv… view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative comparison on editing tasks. [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative comparison on generation task. [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Results of Human Perceptual Evaluation. and further analyze the impact of each module through ablation studies. Previ￾ous methods such as AnyText and AnyText2 show performance improvements after fine-tuning on the InnoText-30K dataset. However, their results still remain inferior to our method. Among all modules, the Font Size-Aware Modulation (FSAM) contributes the most, as its removal increases the LPIPS… view at source ↗
Figure 9
Figure 9. Figure 9: Representative failure cases. robustness of InnoText in generating high-fidelity visual text that aligns with human perceptual expectations in complex real-world environments. 5.6 Failure Cases Studies We analyze representative failure cases, as illustrated in [PITH_FULL_IMAGE:figures/full_fig_p014_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: More human study. the effectiveness of our method in producing visually pleasing and semantically correct outputs. 3 The Prompts Used to Generate Images In our experiments, both visual text editing and generation tasks are performed using fixed prompt templates. For the editing task, the target text is inserted into the template, while for the visual text generation task, image captions are incorporated i… view at source ↗
Figure 11
Figure 11. Figure 11: Qualitative comparison on visual text editing task [PITH_FULL_IMAGE:figures/full_fig_p022_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Qualitative comparison on visual text generation task [PITH_FULL_IMAGE:figures/full_fig_p023_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: The generated image and the corresponding caption content of the image [PITH_FULL_IMAGE:figures/full_fig_p024_13.png] view at source ↗

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

45 extracted references · 16 linked inside Pith

  1. [1]

    In: Proceedings of the AAAI Conference on Artificial Intelligence

    Cao, K., Wang, J., Ma, A., Feng, J., He, X., Ling, R., Liu, H., Lu, J., Feng, W., Wang, H., et al.: Relactrl: Relevance-guided efficient control for diffusion trans- formers. In: Proceedings of the AAAI Conference on Artificial Intelligence. vol. 40, pp. 2598–2606 (2026)

    2026

  2. [2]

    Advances in Neural Information Processing Systems36, 9353–9387 (2023)

    Chen, J., Huang, Y., Lv, T., Cui, L., Chen, Q., Wei, F.: Textdiffuser: Diffusion models as text painters. Advances in Neural Information Processing Systems36, 9353–9387 (2023)

    2023

  3. [3]

    In: European Conference on Computer Vision

    Chen, J., Huang, Y., Lv, T., Cui, L., Chen, Q., Wei, F.: Textdiffuser-2: Unleashing the power of language models for text rendering. In: European Conference on Computer Vision. pp. 386–402. Springer (2024)

    2024

  4. [4]

    arXiv preprint arXiv:2503.23461 (2025)

    Du, N., Chen, Z., Chen, Z., Gao, S., Chen, X., Jiang, Z., Yang, J., Tai, Y.: Textcrafter: Accurately rendering multiple texts in complex visual scenes. arXiv preprint arXiv:2503.23461 (2025)

    arXiv 2025

  5. [5]

    In: Forty-first international conference on machine learning (2024)

    Esser, P., Kulal, S., Blattmann, A., Entezari, R., Müller, J., Saini, H., Levi, Y., Lorenz, D., Sauer, A., Boesel, F., et al.: Scaling rectified flow transformers for high-resolution image synthesis. In: Forty-first international conference on machine learning (2024)

    2024

  6. [6]

    arXiv preprint arXiv:2503.07703 (2025)

    Gong, L., Hou, X., Li, F., Li, L., Lian, X., Liu, F., Liu, L., Liu, W., Lu, W., Shi, Y., et al.: Seedream 2.0: A native chinese-english bilingual image generation foundation model. arXiv preprint arXiv:2503.07703 (2025)

    Pith/arXiv arXiv 2025

  7. [7]

    arXiv preprint arXiv:2502.10999 (2025)

    Jiang, B., Yuan, Y., Bai, X., Hao, Z., Yin, A., Hu, Y., Liao, W., Ungar, L., Tay- lor, C.J.: Controltext: Unlocking controllable fonts in multilingual text rendering without font annotations. arXiv preprint arXiv:2502.10999 (2025)

    arXiv 2025

  8. [8]

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

    2024

  9. [9]

    arXiv preprint arXiv:2505.03329 (2025)

    Lan, R., Bai, Y., Duan, X., Li, M., Sun, L., Chu, X.: Flux-text: A simple and advanced diffusion transformer baseline for scene text editing. arXiv preprint arXiv:2505.03329 (2025)

    arXiv 2025

  10. [10]

    arXiv preprint arXiv:2507.01926 (2025)

    Li, Y., Li, X., Zhang, Z., Bian, Y., Liu, G., Li, X., Xu, J., Hu, W., Liu, Y., Li, L., et al.: Ic-custom: Diverse image customization via in-context learning. arXiv preprint arXiv:2507.01926 (2025)

    arXiv 2025

  11. [11]

    In: Proceedings of the AAAI Conference on Artificial Intelligence

    Ling, R., Cao, K., Lu, J., Ma, A., Liu, H., He, R., Wang, C., Xu, R., Shao, Y., Zhang, Z., et al.: Mofu: Scale-aware modulation and fourier fusion for multi-subject video generation. In: Proceedings of the AAAI Conference on Artificial Intelligence. vol. 40, pp. 7033–7041 (2026)

    2026

  12. [12]

    arXiv preprint arXiv:2209.03003 (2022)

    Liu, X., Gong, C., Liu, Q.: Flow straight and fast: Learning to generate and transfer data with rectified flow. arXiv preprint arXiv:2209.03003 (2022)

    Pith/arXiv arXiv 2022

  13. [13]

    In: European Confer- ence on Computer Vision

    Liu, Z., Liang, W., Liang, Z., Luo, C., Li, J., Huang, G., Yuan, Y.: Glyph-byt5: A customized text encoder for accurate visual text rendering. In: European Confer- ence on Computer Vision. pp. 361–377. Springer (2024) 16 H.Liu et al

    2024

  14. [14]

    arXiv preprint arXiv:2505.24417 (2025)

    Lu, R., Zhang, Y., Liu, J., Wang, H., Song, Y.: Easytext: Controllable diffu- sion transformer for multilingual text rendering. arXiv preprint arXiv:2505.24417 (2025)

    arXiv 2025

  15. [15]

    In: Proceed- ings of the IEEE/CVF International Conference on Computer Vision

    Ma, A., Feng, J., Cao, K., Wang, J., Wang, Y., Zhang, Q., Zhang, Z.: Lay2story: extending diffusion transformers for layout-togglable story generation. In: Proceed- ings of the IEEE/CVF International Conference on Computer Vision. pp. 16102– 16111 (2025)

    2025

  16. [16]

    arXiv preprint arXiv:2412.17225 (2024)

    Ma, L., Yue, T., Fu, P., Zhong, Y., Zhou, K., Wei, X., Hu, J.: Chargen: High accurate character-level visual text generation model with multimodal encoder. arXiv preprint arXiv:2412.17225 (2024)

    Pith/arXiv arXiv 2024

  17. [17]

    completely blind

    Mittal, A., Soundararajan, R., Bovik, A.C.: Making a “completely blind” image quality analyzer. IEEE Signal processing letters20(3), 209–212 (2012)

    2012

  18. [18]

    In: Proceedings of the AAAI conference on artificial intelligence

    Mou, C., Wang, X., Xie, L., Wu, Y., Zhang, J., Qi, Z., Shan, Y.: T2i-adapter: Learning adapters to dig out more controllable ability for text-to-image diffusion models. In: Proceedings of the AAAI conference on artificial intelligence. vol. 38, pp. 4296–4304 (2024)

    2024

  19. [19]

    Peebles,W.,Xie,S.:Scalablediffusionmodelswithtransformers.In:Proceedingsof the IEEE/CVF international conference on computer vision. pp. 4195–4205 (2023)

    2023

  20. [20]

    arXiv preprint arXiv:2307.01952 (2023)

    Podell, D., English, Z., Lacey, K., Blattmann, A., Dockhorn, T., Müller, J., Penna, J., Rombach, R.: Sdxl: Improving latent diffusion models for high-resolution image synthesis. arXiv preprint arXiv:2307.01952 (2023)

    Pith/arXiv arXiv 2023

  21. [21]

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

    Qin, Y., Cao, K., Liu, H., Ma, A., Li, F., Zhu, H., Zhang, Z., Ling, R., Feng, W., He, X., et al.: Innoads-composer: Efficient condition composition for e-commerce poster generation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 32988–32999 (2026)

    2026

  22. [22]

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

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B.: High-resolution image synthesis with latent diffusion models. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 10684–10695 (2022)

    2022

  23. [23]

    arXiv preprint arXiv:2504.15009 (2025)

    Song, W., Jiang, H., Yang, Z., Quan, R., Yang, Y.: Insert anything: Image insertion via in-context editing in dit. arXiv preprint arXiv:2504.15009 (2025)

    Pith/arXiv arXiv 2025

  24. [24]

    IEEE transactions on image processing27(8), 3998–4011 (2018)

    Talebi, H., Milanfar, P.: Nima: Neural image assessment. IEEE transactions on image processing27(8), 3998–4011 (2018)

    2018

  25. [25]

    arXiv preprint arXiv:2411.15098 (2024)

    Tan, Z., Liu, S., Yang, X., Xue, Q., Wang, X.: Ominicontrol: Minimal and universal control for diffusion transformer. arXiv preprint arXiv:2411.15098 (2024)

    Pith/arXiv arXiv 2024

  26. [26]

    arXiv preprint arXiv:2411.15245 (2024)

    Tuo, Y., Geng, Y., Bo, L.: Anytext2: Visual text generation and editing with customizable attributes. arXiv preprint arXiv:2411.15245 (2024)

    Pith/arXiv arXiv 2024

  27. [27]

    arXiv preprint arXiv:2311.03054 (2023)

    Tuo, Y., Xiang, W., He, J.Y., Geng, Y., Xie, X.: Anytext: Multilingual visual text generation and editing. arXiv preprint arXiv:2311.03054 (2023)

    Pith/arXiv arXiv 2023

  28. [28]

    arXiv preprint arXiv:2504.19724 (2025)

    Wang, H., Xu, Y., Li, Y., Li, J., Zhang, C., Wang, J., Yang, K., Chen, Z.: Reptext: Rendering visual text via replicating. arXiv preprint arXiv:2504.19724 (2025)

    Pith/arXiv arXiv 2025

  29. [29]

    Ad- vances in Neural Information Processing Systems38, 5388–5416 (2026)

    Wang, J., Ma, A., Cao, K., Zheng, J., Feng, J., Zhang, Z., Pang, W., Liang, X.: Wisa: World simulator assistant for physics-aware text-to-video generation. Ad- vances in Neural Information Processing Systems38, 5388–5416 (2026)

    2026

  30. [30]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Wang,Y.,Zhang,W.,Xu,H.,Jin,C.:Dreamtext:Highfidelityscenetextsynthesis. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 28555–28563 (2025)

    2025

  31. [31]

    arXiv preprint arXiv:2412.10302 (2024) InnoText: A Unified Model for Visual Text Generation and Editing 17

    Wu,Z.,Chen,X.,Pan,Z.,Liu,X.,Liu,W.,Dai,D.,Gao,H.,Ma,Y.,Wu,C.,Wang, B., et al.: Deepseek-vl2: Mixture-of-experts vision-language models for advanced multimodal understanding. arXiv preprint arXiv:2412.10302 (2024) InnoText: A Unified Model for Visual Text Generation and Editing 17

    Pith/arXiv arXiv 2024

  32. [32]

    arXiv preprint arXiv:2505.17778 (2025)

    Xie, Y., Zhang, J., Chen, P., Wang, Z., Wang, W., Gao, L., Li, P., Sun, H., Zhang, Q., Qiao, Q., et al.: Textflux: An ocr-free dit model for high-fidelity multilingual scene text synthesis. arXiv preprint arXiv:2505.17778 (2025)

    arXiv 2025

  33. [33]

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

    Xu, Y., Feng, W., Zhang, S., Wang, H., Qin, Y., Li, Y., Ma, A., Luo, Y., Wang, L., Ren, X., et al.: Design your ad: Personalized advertising image and text generation with unified autoregressive models. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 472–483 (2026)

    2026

  34. [34]

    Advances in Neural Informa- tion Processing Systems36, 44050–44066 (2023)

    Yang, Y., Gui, D., Yuan, Y., Liang, W., Ding, H., Hu, H., Chen, K.: Glyphcontrol: Glyph conditional control for visual text generation. Advances in Neural Informa- tion Processing Systems36, 44050–44066 (2023)

    2023

  35. [35]

    arXiv preprint arXiv:2308.06721 (2023)

    Ye, H., Zhang, J., Liu, S., Han, X., Yang, W.: Ip-adapter: Text compati- ble image prompt adapter for text-to-image diffusion models. arXiv preprint arXiv:2308.06721 (2023)

    Pith/arXiv arXiv 2023

  36. [36]

    Advances in Neural Information Processing Systems37, 138569–138594 (2024)

    Zeng, W., Shu, Y., Li, Z., Yang, D., Zhou, Y.: Textctrl: Diffusion-based scene text editing with prior guidance control. Advances in Neural Information Processing Systems37, 138569–138594 (2024)

    2024

  37. [37]

    Zhang,B.,Gao,Z.,Qu,Y.,Xie,H.:Howcontrolinformationinfluencesmultilingual text image generation and editing? Advances in Neural Information Processing Systems37, 6884–6904 (2024)

    2024

  38. [38]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Zhang, L., Rao, A., Agrawala, M.: Adding conditional control to text-to-image diffusion models. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 3836–3847 (2023)

    2023

  39. [39]

    arXiv preprint arXiv:2503.07027 (2025)

    Zhang, Y., Yuan, Y., Song, Y., Wang, H., Liu, J.: Easycontrol: Adding efficient and flexible control for diffusion transformer. arXiv preprint arXiv:2503.07027 (2025)

    Pith/arXiv arXiv 2025

  40. [40]

    arXiv preprint arXiv:2504.20690 (2025)

    Zhang, Z., Xie, J., Lu, Y., Yang, Z., Yang, Y.: In-context edit: Enabling instruc- tional image editing with in-context generation in large scale diffusion transformer. arXiv preprint arXiv:2504.20690 (2025)

    Pith/arXiv arXiv 2025

  41. [41]

    arXiv preprint arXiv:2503.08157 (2025)

    Zhang, Z., Ma, A., Cao, K., Wang, J., Liu, S., Ma, Y., Cheng, B., Leng, D., Yin, Y.: U-stydit: Ultra-high quality artistic style transfer using diffusion transformers. arXiv preprint arXiv:2503.08157 (2025)

    Pith/arXiv arXiv 2025

  42. [42]

    arXiv preprint arXiv:2503.21749 (2025)

    Zhao, S., Wu, Q., Li, X., Zhang, B., Li, M., Qin, Q., Liu, D., Zhang, K., Li, H., Qiao, Y., et al.: Lex-art: Rethinking text generation via scalable high-quality data synthesis. arXiv preprint arXiv:2503.21749 (2025)

    Pith/arXiv arXiv 2025

  43. [43]

    In: European conference on computer vision

    Zhao, Y., Lian, Z.: Udifftext: A unified framework for high-quality text synthesis in arbitrary images via character-aware diffusion models. In: European conference on computer vision. pp. 217–233. Springer (2024)

    2024

  44. [44]

    arXiv preprint arXiv:2504.10479 (2025)

    Zhu, J., Wang, W., Chen, Z., Liu, Z., Ye, S., Gu, L., Tian, H., Duan, Y., Su, W., Shao, J., et al.: Internvl3: Exploring advanced training and test-time recipes for open-source multimodal models. arXiv preprint arXiv:2504.10479 (2025)

    Pith/arXiv arXiv 2025

  45. [45]

    HARRY POTTED PLANT ,\

    Zhu,Y.,Liu,J.,Gao,F.,Liu,W.,Wang,X.,Wang,P.,Huang,F.,Yao,C.,Yang,Z.: Visual text generation in the wild. In: European Conference on Computer Vision. pp. 89–106. Springer (2024) InnoText: A Unified Model for Visual Text Generation and Editing 1 1 More Quantitative Results In addition to evaluating the visual text editing capabilities of various methods usi...

    arXiv 2024