Research map / Image generation / Text-to-image diffusion
Long-Text-to-Image Generation via Compositional Prompt Decomposition
Jen-Yuan Huang, Tong Lin, Yilun Du
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add PRISM compositional decomposition to Qwen-Image | Qwen-Image · DetailMaster | Character Presence accuracy: 40.46 -> 46.84 | Improved |
| add PRISM to Qwen-Image | Qwen-Image · DetailMaster long prompts | HPSv3: 8.56 -> 12.05 | Improved |
| add PRISM decomposition vs fine-tuned LongAlign on prompts over 500 tokens | PRISM-SD1.5 · DetailMaster prompts >500 tokens | accuracy: 7.4% | Improved |
| combine PRISM with LongAlign reward tuning | PRISM-SD1.5 · DetailMaster | Spatial Relation accuracy: 20.88 -> 24.47 | Improved |
| increase number of components N from 3 to 5 (ablation) | PRISM-SD1.5 · DetailMaster | gain over compute-matched fine-tuning | Improved |
| add multiple components instead of a single component (non-compositional) (ablation) | PRISM-SD1.5 · DetailMaster | accuracy | Improved |
| learned decomposition instead of simple sentence splitting (ablation) | PRISM-SD1.5 · DetailMaster | accuracy | Improved |
| add PRISM LoRA adaptation to FLUX-Dev. | FLUX-Dev. · DetailMaster (appendix) | Character Presence accuracy: 34.33 -> 31.05 | Hurt |
| add PRISM LoRA adaptation to SD-3.5 Medium | SD-3.5 Medium · DetailMaster (appendix) | Character Presence accuracy: 31.19 -> 33.03 | Improved |
| add PRISM to SD-1.5 (vs LongAlign) on standard-length prompts | PRISM-SD1.5 · GenEval | GenEval score: 0.5075 -> 0.4960 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- LLM4GEN — LLM4GEN: Leveraging Semantic Representation of LLMs for Text-to-Image Generation (AAAI 2025) · better on 1 of 1 shared result
- LLM Blueprint — LLM Blueprint: Enabling Text-to-Image Generation with Complex and Detailed Prompts (ICLR 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- LLM Blueprint: Enabling Text-to-Image Generation with Complex and Detailed Prompts (ICLR 2024 · similarity 0.90)
- A User-Friendly Framework for Generating Model-Preferred Prompts in Text-to-Image Synthesis (AAAI 2024 · similarity 0.88)
- Compositional Text-to-Image Generation with Dense Blob Representations (ICML 2024 · similarity 0.88)
- ConceptMix: A Compositional Image Generation Benchmark with Controllable Difficulty (NeurIPS 2024 · similarity 0.88)
- Compositional Text-to-Image Synthesis with Attention Map Control of Diffusion Models (AAAI 2024 · similarity 0.88)
- Visually Guided Decoding: Gradient-Free Hard Prompt Inversion with Language Models (ICLR 2025 · similarity 0.88)
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.
