Research map / Image generation / Text-to-image diffusion
POSTA: A Go-to Framework for Customized Artistic Poster Generation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Presents POSTA, a modular diffusion + MLLM framework that generates customized artistic posters (background, layout/typography via Design MLLM, and stylized title text), and introduces the PosterArt dataset.
Problem
Automatic poster generation lacks text accuracy, user customization, and aesthetic quality needed for artistic posters (movies, exhibitions, concerts).
Approach
Decomposes poster creation into three modules: FLUX-LoRA background diffusion, an LLaVA-based Design MLLM that predicts editable vector typography (rendered directly, not generated as pixels), and a BrushNet-based ArtText Diffusion that stylizes title text via mask-guided inpainting.
Key results
- In the 60-user + GPT-4V evaluation, POSTA outperforms FLUX-dev, Recraft V3, Ideogram v2, SD3 and Playground-v3 across aesthetics, text readability and prompt relevance (exact numbers shown only in a figure).
- Word-level OCR precision and recall (GOT-OCR) are highest for POSTA, which renders text directly from predicted attributes.
- Design MLLM is LLaVA-based with Llama3 7B and ViT-L/14, trained on 2 A800 GPUs.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use POSTA instead of the compared poster-generation models | Poster generation · 20 poster prompts | text consistency with user input | Helped |
| Use POSTA instead of the compared poster-generation models | Poster generation · 20 poster prompts | aesthetics, text readability, and prompt relevance | Helped |
Limitations
- Design MLLM generates only relatively simple layouts and offers a restricted set of font types, limiting design diversity.
- ArtText Diffusion performance is constrained by the limited size of high-standard artistic training data.
- Evaluation relies on a small set of 20 prompts and on human/GPT-4V ratings.
Experimental setup
Datasets: PosterArt-Design PosterArt-Text
Benchmarks: PosterArt-based qualitative comparisons human and GPT-4V poster ratings (20 prompts) GOT-OCR text precision/recall
Metrics: visual appeal/aesthetics rating (1-10) text readability prompt relevance word-level precision word-level recall
Baselines: COLE OpenCOLE AnyText TextDiffuser-2 FLUX-dev SD3 Playground-v3 Recraft V3 Ideogram v2
Builds on: FLUX LLaVA BrushNet COLE AnyText
Closest papers by meaning
- GlyphDraw2: Automatic Generation of Complex Glyph Posters with Diffusion Models and Large Language Models (AAAI 2025 · similarity 0.88)
- PosterO: Structuring Layout Trees to Enable Language Models in Generalized Content-Aware Layout Generation (CVPR 2025 · similarity 0.87)
- Paper2Poster: Towards Multimodal Poster Automation from Scientific Papers (NeurIPS 2025 · similarity 0.87)
- P2P: Automated Paper-to-Poster Generation and Fine-Grained Benchmark (ICLR 2026 · similarity 0.86)
- DesignDiffusion: High-Quality Text-to-Design Image Generation with Diffusion Models (CVPR 2025 · similarity 0.85)
- Conditional Balance: Improving Multi-Conditioning Trade-Offs in Image Generation (CVPR 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.