Research map / Image generation / Text-to-image diffusion
Context Diffusion: In-Context Aware Image Generation
Ivona Najdenkoska, Animesh Sinha, Abhimanyu Dubey, Dhruv Mahajan, Vignesh Ramanathan, Filip Radenovic
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 |
|---|---|---|---|
| inject visual context via cross-attention alongside text instead of Prompt Diffusion-style conditioning | Context Diffusion · Human eval in-domain, context only | win rate (%): 4.5 -> 80.2 | Improved |
| inject visual context via cross-attention vs Prompt Diffusion on out-of-domain tasks | Context Diffusion · Human eval out-of-domain, context+prompt | win rate (%): 26.9 -> 52.3 | Improved |
| prompt-only conditioning on in-domain tasks | Context Diffusion · Human eval in-domain, prompt only | win rate (%): 30.4 -> 29.6 | No clear effect |
| prompt-only conditioning on out-of-domain tasks | Context Diffusion · Human eval out-of-domain, prompt only | win rate (%): 25.9 -> 49.8 | Improved |
| use three context images instead of one (ablation) | Context Diffusion · Human eval out-of-domain few-shot, context only | win rate (%): 28.3 -> 50.2 | Improved |
| use three context images instead of one with prompt (ablation) | Context Diffusion · Human eval out-of-domain few-shot, context+prompt | win rate (%): 21.5 -> 60.0 | Improved |
| pair source image with target as context (as in Prompt Diffusion) (ablation) | Context Diffusion · qualitative | image quality | No clear effect |
| encode context with ConvNets summed to the query image instead of visual conditioning in cross-attention (ablation) | Context Diffusion · qualitative | context fidelity | Hurt |
| context-only conditioning for depth-to-image | Context Diffusion · depth-to-image, context only | FID: 27.0 -> 18.5 | Improved |
| prompt-only conditioning for seg-to-image | Context Diffusion · seg-to-image, prompt only | FID: 17.2 -> 18.5 | Hurt |
Closest papers by meaning
- Prompt-Free Diffusion: Taking "Text" out of Text-to-Image Diffusion Models (CVPR 2024 · similarity 0.89)
- R&B: Region and Boundary Aware Zero-shot Grounded Text-to-image Generation (ICLR 2024 · similarity 0.88)
- Devil is in the Detail: Towards Injecting Fine Details of Image Prompt in Image Generation via Conflict-free Guidance and Stratified Attention (CVPR 2025 · similarity 0.88)
- Explore In-Context Segmentation via Latent Diffusion Models (AAAI 2025 · similarity 0.87)
- CONFORM: Contrast is All You Need for High-Fidelity Text-to-Image Diffusion Models (CVPR 2024 · similarity 0.87)
- Prompting Hard or Hardly Prompting: Prompt Inversion for Text-to-Image Diffusion Models (CVPR 2024 · similarity 0.87)
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.
