PosterCraft - An aesthetic poster generation framework launched by HKUST in collaboration with Meituan and others.
PosterCraft is a unified framework developed by the Hong Kong University of Science and Technology (Guangzhou) and Meituan, among other institutions, for generating high-quality, aesthetically pleasing posters. The framework abandons modular design processes and fixed predefined layouts, supporting free exploration and coherence of models...
What is PosterCraft?
PosterCraft is a unified framework developed by the Hong Kong University of Science and Technology (Guangzhou) and Meituan, among other institutions, for generating high-quality, aesthetically pleasing posters. The framework abandons modular design processes and fixed predefined layouts, allowing models to freely explore coherent and visually appealing compositions. It optimizes the generation of aesthetically pleasing posters based on a cascading workflow of four key stages: scalable text rendering optimization, high-quality poster fine-tuning, reinforcement learning based on aesthetic text, and visual-linguistic feedback refinement. Each stage is supported by a dedicated automated data construction pipeline to meet specific needs, enabling robust training without complex architectural modifications. In multiple experiments, PosterCraft significantly outperforms open-source baselines in rendering accuracy, layout coherence, and overall visual appeal, approaching the quality of commercial systems.
Main functions of PosterCraft
- High-quality text renderingIt can accurately render text, ensuring the clarity and accuracy of the text content.
- Artistic Content CreationGenerate visual content with an abstract artistic feel, giving posters a unique aesthetic style.
- Eye-catching layout designCreate a visually impactful layout to ensure overall design harmony and consistency.
- End-to-end generationFrom text input to the generation of the final poster, the entire process is completed in a single model, without the need for external modules or preset templates.
- Aesthetic optimizationBased on reinforcement learning and visual-language feedback mechanisms, the aesthetic quality and content accuracy of posters are optimized.
The technical principles of PosterCraft
- Scalable text rendering optimizationBased on the large-scale Text-Render-2M dataset, we optimize the training of the model for text rendering, thereby improving the accuracy and clarity of the text.
- High-quality poster fine-tuningSupervised fine-tuning based on the HQ-Poster-100K dataset improves the overall visual quality and consistency of artistic style of the posters.
- Reinforcement learning based on aesthetic textsUsing the Poster-Preference-100K dataset, we trained a model based on optimal preference optimization to generate posters that are more in line with human aesthetics.
- Visual-verbal feedback refinementBy leveraging the Poster-Reflect-120K dataset and incorporating multimodal feedback, the generated posters were further optimized and adjusted to improve the accuracy and aesthetic value of the content.
PosterCraft's project address
- Project official websitehttps://ephemeral182.github.io/PosterCraft/
- GitHub repository: https://github.com/Ephemeral182/PosterCraft
- HuggingFace model libraryhttps://huggingface.co/PosterCraft
- arXiv technical paper: https://arxiv.org/pdf/2506.10741
Application scenarios of PosterCraft
- Movie posterGenerate attractive posters based on the movie's theme, highlighting key elements and visual impact.
- Art exhibition posterGenerate posters with an artistic and cultural feel, showcasing the exhibition's concept and style.
- Product promotion posterGenerate attractive promotional posters based on product features to showcase functionality and advantages.
- Academic conference posterGenerate posters that convey a professional and academic atmosphere, showcasing the conference theme and agenda.
- Campus Activity PostersCreate creative posters to showcase the event's content and highlights.