Research map / Image generation
Text-to-image diffusion: research map
1,916 papers on Text-to-image diffusion in Image generation, from ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR (2016–2026), grouped into 3 clusters and 12 approaches. The busiest year so far is 2026.
What it is
Text-to-image diffusion models (the family behind Stable Diffusion) turn a text prompt into an image. Research makes them follow prompts more faithfully, place objects where asked, personalise them to a subject from a few photos, align them with human preferences and remove unwanted concepts.
Terms you will see on this page:
- T2I
- text-to-image generation.
- Personalization
- teaching the model a specific subject or style from a few images.
- Concept erasure
- removing a concept (an artist's style, unsafe content) from a trained model.
Start reading with the most cited papers here:
- DreamBooth: Fine Tuning Text-to-Image Diffusion Models for Subject-Driven Generation (CVPR 2023 · 2,183 citations)
- Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding (NeurIPS 2022 · 2,090 citations)
Within Image generation, its share grew from 18.8% in 2023–24 to 20.8% in 2025–26 (391 → 851 papers at ICML, NeurIPS, ICLR, CVPR and AAAI, the venues with data for all four years).
Explore Text-to-image diffusion in the interactive map
Working on something in this topic? Describe your idea in scime atlas to see which approach it falls under, the closest papers by meaning and how crowded the spot has become.
What changed recently
Approaches whose share of the topic moved most between 2023–24 and 2025–26 (ICML, NeurIPS, ICLR, CVPR and AAAI):
- T2I evaluation and controllable adapters rising: 7% → 14% of the topic (26 → 121 papers)
- Preference optimization for diffusion rising: 2% → 9% of the topic (8 → 74 papers)
- Layout-to-image control rising: 9% → 15% of the topic (36 → 125 papers)
- Compositional text-to-image guidance falling: 22% → 12% of the topic (87 → 105 papers)
- Text-driven synthesis and semantic guidance falling: 15% → 9% of the topic (60 → 77 papers)
Papers from 2025–2026 that are already cited most:
- OmniGen: Unified Image Generation (CVPR 2025 · 56 citations)
- VideoDPO: Omni-Preference Alignment for Video Diffusion Generation (CVPR 2025 · 18 citations)
Approaches and key papers
Composition, layout and personalization · 1,245 papers
Keywords: synthesis, layout, object
Approaches in this cluster:
- Compositional text-to-image guidance (300 papers)
Improve object composition in text-to-image diffusion through attention guidance and token-level supervision. - Text-driven synthesis and semantic guidance (223 papers)
Control text-to-image models for size, semantics, and LLM-guided correction. - Personalized and region-controlled generation (244 papers)
Generate subject-driven images with personalization and region control, often without fine-tuning. - Layout-to-image control (259 papers)
Control spatial arrangement with layouts, instance maps, and spatio-textual representations. - Multi-subject personalization (133 papers)
Preserve identity and disentangle attributes for personalized multi-subject generation. - Text generation and structured prediction (86 papers)
Generate text or images with structured prediction and controlled decoding.
Most cited and most cited since 2024:
- DreamBooth: Fine Tuning Text-to-Image Diffusion Models for Subject-Driven Generation (CVPR 2023 · 2,183 citations)
- AttnGAN: Fine-Grained Text to Image Generation With Attentional Generative Adversarial Networks (CVPR 2018 · 1,935 citations)
- SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis (ICLR 2024 · 322 citations)
- Style Aligned Image Generation via Shared Attention (CVPR 2024 · 131 citations)
Alignment and evaluation of T2I models · 502 papers
Keywords: t2i, preference, alignment
Approaches in this cluster:
- Inference-time T2I improvement (131 papers)
Improve fidelity, safety, and diversity of text-to-image models through noise optimization and guidance. - T2I evaluation and controllable adapters (192 papers)
Benchmark text-to-image models with human feedback and add controllable adapters. - Preference optimization for diffusion (110 papers)
Align diffusion models with human preferences using direct preference optimization. - Fairness and bias in generative models (69 papers)
Measure and mitigate bias in text-to-image generation.
Most cited and most cited since 2024:
- Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding (NeurIPS 2022 · 2,090 citations)
- T2I-Adapter: Learning Adapters to Dig Out More Controllable Ability for Text-to-Image Diffusion Models (AAAI 2024 · 865 citations)
- Rethinking FID: Towards a Better Evaluation Metric for Image Generation (CVPR 2024 · 200 citations)
- Rich Human Feedback for Text-to-Image Generation (CVPR 2024 · 63 citations)
Concept erasure · 169 papers
Keywords: concept, erasure, unlearning
Approaches in this cluster:
- Concept erasure in diffusion models (103 papers)
Remove unwanted concepts from diffusion models via editing, adapters, and unlearning. - Multi-concept composition (66 papers)
Customize and compose multiple concepts in text-to-image diffusion models.
Most cited and most cited since 2024:
- Multi-Concept Customization of Text-to-Image Diffusion (CVPR 2023 · 627 citations)
- MACE: Mass Concept Erasure in Diffusion Models (CVPR 2024 · 73 citations)
- Self-Discovering Interpretable Diffusion Latent Directions for Responsible Text-to-Image Generation (CVPR 2024 · 30 citations)
- One-dimensional Adapter to Rule Them All: Concepts Diffusion Models and Erasing Applications (CVPR 2024 · 28 citations)
Related topics in Image generation
- Autoregressive and latent generation (2,100)
- GANs (780)
- Super-resolution (771)
- Editing and style transfer (1,430)
- Fast diffusion sampling (2,155)
- Image restoration (2,009)
