Research map / Image generation / Text-to-image diffusion
ITI-GEN: Inclusive Text-to-Image Generation
Cheng Zhang, Xuanbai Chen, Siqi Chai, Chen Henry Wu, Dmitry Lagun, Thabo Beeler, Fernando De la Torre
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 |
|---|---|---|---|
| learn inclusive tokens from reference images instead of hard prompt specification | Stable Diffusion · CelebA eyeglass | KL divergence: 0.371 -> 2 × 10^-4 | Improved |
| aggregate separate inclusive tokens by element-wise sum for multiple attributes | ITI-GEN · CelebA male x young x eyeglass x smile | KL divergence: 0.476 -> 0.094 | Improved |
| add semantic consistency loss L_sem (ablation) | ITI-GEN (CelebA reference) · FID on FFHQ | FID: 77.78 -> 60.38 | Improved |
| add semantic consistency loss L_sem (ablation) | ITI-GEN (FAIR reference) · FID on FFHQ | FID: 62.69 -> 51.83 | Improved |
| use synthetic FAIR reference images instead of real CelebA images (ablation) | ITI-GEN · FID on FFHQ | FID: 60.38 -> 51.83 | Improved |
| increase reference images per category beyond 10 (ablation) | ITI-GEN · quantity ablation | quality | Improved |
| use Custom Diffusion (personalization) with reference images instead of ITI-GEN | Stable Diffusion · CelebA smile | KL divergence: 0.134 -> 0.469 | Hurt |
| apply ethical intervention prompts | Stable Diffusion · CelebA pale skin | KL divergence: 0.308 -> 0.644 | Hurt |
Later papers that compared with this method
5 later accepted papers report ITI-GEN as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.
- AITTI: Learning Adaptive Inclusive Token for Text-to-Image Generation (IJCV 2026 · better on 1 of 2 shared results)
- FairGen: Enhancing Fairness in Text-to-Image Diffusion Models via Self-Discovering Latent Directions (ICCV 2025 · better on 5 of 5 shared results)
- Multi-Group Proportional Representations for Text-to-Image Models (CVPR 2025 · better on 3 of 3 shared results)
- FairQueue: Rethinking Prompt Learning for Fair Text-to-Image Generation (NeurIPS 2024 · better on 14 of 16 shared results)
- Training Diffusion Models Towards Diverse Image Generation with Reinforcement Learning (CVPR 2024 · better on 4 of 6 shared results)
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
- LLM4GEN: Leveraging Semantic Representation of LLMs for Text-to-Image Generation (AAAI 2025 · similarity 0.90)
- LLM Blueprint: Enabling Text-to-Image Generation with Complex and Detailed Prompts (ICLR 2024 · similarity 0.89)
- Salient Concept-Aware Generative Data Augmentation (NeurIPS 2025 · similarity 0.89)
- A User-Friendly Framework for Generating Model-Preferred Prompts in Text-to-Image Synthesis (AAAI 2024 · similarity 0.88)
- DreamDistribution: Learning Prompt Distribution for Diverse In-distribution Generation (ICLR 2025 · similarity 0.88)
- Rich Human Feedback for Text-to-Image Generation (CVPR 2024 · 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.
