Research map / Image generation / Text-to-image diffusion
TIPO: Text to Image with Text Pre-sampling for Prompt Optimization
Shih-Ying Yeh, YI LI, SangHyun Park, Giyeong Oh, Xuehai Wang, Min Song, Youngjae Yu, Shang-Hong Lai
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 |
|---|---|---|---|
| expand tag prompts with TIPO | Kohaku-XL-Zeta · in-domain tag prompts | FDD: 0.3558 -> 0.2282 | Improved |
| expand tag prompts with TIPO | Kohaku-XL-Zeta · in-domain tag prompts | AI Corrupt score: 0.5743 -> 0.0805 | Improved |
| expand short NL prompts with TIPO | SDXL-base-1.0 · in-domain NL-short | FDD: 0.0957 -> 0.1168 | Hurt |
| expand short NL prompts with TIPO | SDXL-base-1.0 · in-domain NL-short | Aesthetic score: 5.8370 -> 5.8531 | Improved |
| apply TIPO on out-of-domain prompts | SD-3.5-Large · out-of-domain | Vendi score: 8.9718 -> 21.571 | Improved |
| apply TIPO instead of GPT-4o-mini rewriting on out-of-domain prompts | SD-3.5-Large · out-of-domain | Aesthetic score: 6.7125 -> 6.0536 | Hurt |
| apply TIPO on models with undisclosed training data | FLUX.1-dev, Lumina-2, HiDream-I1 · COYO/GBC 1000 prompts | Aesthetic score: 5.2588 -> 5.4105 (Lumina-2) | Improved |
| apply TIPO on Omnigen2 | Omnigen2 · COYO/GBC 1000 prompts | Aesthetic score: 5.2629 -> 5.0661 | Hurt |
| use TIPO for prompt generation instead of Promptist | TIPO · prompt latency | avg. time (s): 1.46 -> 1.03 | Improved |
| use TIPO prompts instead of baselines in human study | TIPO · human preference | win rate (%): 51.3 | Improved |
| use TIPO instead of MagicPrompt on NL prompts | TIPO · CLIPScore prompt-image, NL-short | CLIPScore: 0.2834 -> 0.2413 | Hurt |
| use TIPO prompts for embedding alignment with T2I training text | TIPO · prompt distribution alignment | Frechet distance: 0.2578 -> 0.0524 (T5, tag-based vs MagicPrompt) | Improved |
Closest papers by meaning
- A User-Friendly Framework for Generating Model-Preferred Prompts in Text-to-Image Synthesis (AAAI 2024 · similarity 0.90)
- PromptCoT: Align Prompt Distribution via Adapted Chain-of-Thought (CVPR 2024 · similarity 0.89)
- Prompt-Free Diffusion: Taking "Text" out of Text-to-Image Diffusion Models (CVPR 2024 · similarity 0.89)
- Dynamic Prompt Optimizing for Text-to-Image Generation (CVPR 2024 · similarity 0.89)
- Focus-N-Fix: Region-Aware Fine-Tuning for Text-to-Image Generation (CVPR 2025 · similarity 0.89)
- Multimodal Prompt Optimization: Why Not Leverage Multiple Modalities for MLLMs (ICLR 2026 · similarity 0.89)
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.
