PosterText: Towards Unified Visual Text Generation and Editing for E-commerce Poster

Xiaoan Liu; Lichen Ma; Zipeng Guo; Yu He; Xiaoyan Su; Shaojie Guo; Jingling Fu; Xiaolong Fu

cs.CVAug 17, 2026

PosterText: Towards Unified Visual Text Generation and Editing for E-commerce Poster

Authors: Xiaoan Liu, Lichen Ma, Zipeng Guo, Yu He, Xiaoyan Su, Shaojie Guo, Jingling Fu, Xiaolong Fu,

Organizations: Wuhan University · State Key Laboratory of Human-Machine Hybrid Augmented Intelligence, Institute of Artificial Intelligence and Robotics, Xi’an Jiaotong University · The Hong Kong University of Science and Technology (Guangzhou)

Abstract

Automated e-commerce poster design requires both high-quality poster generation and flexible editing of existing designs. However, most existing methods either target end-to-end poster generation or follow multi-stage design pipelines, with limited capability for flexible and precise editing of existing posters. To enable unified generation and editing of e-commerce posters, we introduce Text Patch Generation and Editing, a unified task formulation that treats text patches as atomic units and covers four operations: poster generation, patch addition, patch deletion, and patch modification, with optional reference-guided style control. Based on this, we propose PosterText, a unified model trained with a four-stage curriculum, including text rendering pretraining, instruction-following training, reinforcement learning for preference alignment, and spatial guidance self-distillation for execution refinement. We further construct a large-scale dataset with patch-level annotations and a comprehensive benchmark for evaluation. Extensive experiments demonstrate that PosterText achieves competitive performance against existing generation and editing approaches, validating the effectiveness of the proposed framework.

Figures & tables

Figure 1: Overview of PosterText framework and qualitative comparison. (a) Four unified operations: poster generation, patch addition, patch deletion, and patch modification. (b) PosterText enables both poster generation and localized editing within a single framework. (c) Qualitative comparison with closed-source baselines.

Figure 2: Overview of PosterText. PosterText is trained through four stages: text patch rendering pretraining, instruction-following training, reinforcement learning with human preferences, and Spatial Guidance Self-Distillation (SGSD). The framework supports four operations: generation, addition, deletion, and modification. For SGSD training, the teacher model uses mask-guided inputs for spatial supervision, while the student learns mask-free localization through on-policy self-distillation.

Figure 3: Overview of the data construction pipeline. The pipeline first extracts patch-level structural annotations from raw posters, and then constructs operation pairs for generation, addition, deletion, and modification.

Figure 4: Visual comparison of PosterText with existing methods.

Closed-Source Methods
	Patch Addition					Patch Deletion			Patch Modification					Poster Generation
Method	S.A.	NED	M-L.	I.F.	V.Q.	M-L.	I.F.	V.Q.	S.A.	NED	M-L.	I.F.	V.Q.	I.F.	V.Q.	B.P.
Seedream 5.0	0.403	0.585	0.195	4.40	4.64	0.122	4.68	4.69	0.735	0.884	0.102	4.33	4.65	4.10	4.50	4.80
Nano Banana 2	0.500	0.635	0.064	4.46	4.67	0.051	4.62	4.84	0.819	0.907	0.034	4.47	4.80	4.63	4.76	4.86
Nano Banana Pro	0.543	0.679	0.184	4.55	4.75	0.061	4.66	4.91	0.815	0.891	0.102	4.36	4.64	4.58	4.88	4.92
Open-Source Methods

Table 1: Quantitative results on PosterText-Bench. Bold and underline indicate the best and second-best results, respectively. S.A.: Sentence Accuracy; M-L.: Masked-LPIPS; I.F.: Instruction Following; V.Q.: Visual Quality; B.P.: Background Preservation. Higher

\uparrow

is better for S.A., NED, I.F., V.Q., and B.P., while lower

\downarrow

is better for M-L.

Model	S.A. $\uparrow$	NED $\uparrow$	M-L. $\downarrow$	I.F. $\uparrow$	V.Q. $\uparrow$
Base Model	0.490	0.666	0.417	3.27	3.37
+SFT Stage
Stage II-only	0.679	0.786	0.088	3.63	3.81
Stage I(no cur.)+II	0.702	0.831	0.061	4.00	4.12
Stage I + II	0.712	0.866	0.053	4.03	4.24
+SFT+RL	0.735	0.883	0.037	4.22	4.51

Table 2: Ablation study of different training stages, averaged across all tasks.

T2I Generation

Text-to-image generation creates images from natural-language prompts using generative models, with research focused on fidelity, control, efficiency, and safety.

See all 291 papers

InnoAds-Composer: Efficient Condition Composition for E-Commerce Poster Generation
Mar 6, 2026Yuxin Qin, Ke Cao, Haowei Liu +13Image Generation Compositional Image Generation
JD.com, Inc., Beijing, China · Chongqing University of Posts and Telecommunications, Chongqing, China · The Hong Kong University of Science and Technology, Hong Kong, China +1
simpleposter: a simple baseline for product poster generation
May 9, 2026Benlei Cui, Fangao Zeng, Weitao Jiang +6Image Generation Image Inpainting
1Alibaba Group · 2Taobao & Tmall Group of Alibaba
TextRefine: Improving Textual Fidelity, Spatial Placement, and Glyph Rendering for Text Editing in Product Posters
Aug 20, 2026Honglie Wang, Jia Sun, Zijun Li +11Image Editing Text-Guided Image Editing
Kuaishou Technology · School of Artificial Intelligence, University of Chinese Academy of Sciences, Beijing 100049, China · State Key Laboratory of Multimodal Artificial Intelligence Systems (MAIS), Institute of Automation, Chinese Academy of Sciences, Beijing 100190, China

Mar 6, 2026J/K move · Enter open · S save

InnoAds-Composer: Efficient Condition Composition for E-Commerce Poster Generation

Yuxin Qin, Ke Cao, Haowei Liu, Ao Ma+12

JD.com, Inc., Beijing, China · Chongqing University of Posts and Telecommunications, Chongqing, China · The Hong Kong University of Science and Technology, Hong Kong, China+1

Image Generation Compositional Image Generation Conditional Diffusion Models

Open paper AskPDF arXiv

E-commerce product poster generation aims to automatically synthesize a single image that effectively conveys product information by presenting a subject, text, and a designed style. Recent diffusion models with fine-grained and efficient controllability have advanced product poster synthesis, yet they typically rely on multi-stage pipelines, and simultaneous control over subject, text, and style remains underexplored. Such naive multi-stage pipelines also show three issues: poor subject fidelity, inaccurate text, and inconsistent style. To address these issues, we propose InnoAds-Composer, a single-stage framework that enables efficient tri-conditional control tokens over subject, glyph, and style. To alleviate the quadratic overhead introduced by naive tri-conditional token concatenation, we perform importance analysis over layers and timesteps and route each condition only to the most responsive positions, thereby shortening the active token sequence. Besides, to improve the accuracy of Chinese text rendering, we design a Text Feature Enhancement Module (TFEM) that integrates features from both glyph images and glyph crops. To support training and evaluation, we also construct a high-quality e-commerce product poster dataset and benchmark, which is the first dataset that jointly contains subject, text, and style conditions. Extensive experiments demonstrate that InnoAds-Composer significantly outperforms existing product poster methods without obviously increasing inference latency.

Abstract

T2I Generation

Explore similar work

InnoAds-Composer: Efficient Condition Composition for E-Commerce Poster Generation

simpleposter: a simple baseline for product poster generation

TextRefine: Improving Textual Fidelity, Spatial Placement, and Glyph Rendering for Text Editing in Product Posters