Research map / Image generation / Text-to-image diffusion
Compositional Visual Generation with Composable Diffusion Models
Nan Liu, Shuang Li, Yilun Du, Antonio Torralba, Joshua B. Tenenbaum
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 |
|---|---|---|---|
| compose single-concept diffusion models with conjunction instead of using EBM composition | Composable Diffusion · CLEVR (1 component) | classification accuracy (%): 86.42 vs 70.54 (EBM); 15.88% higher | Improved |
| compose three position concepts with conjunction instead of EBM | Composable Diffusion · CLEVR (3 components) | classification accuracy (%): 31.36 vs 7.34 (EBM); 24.02% higher | Improved |
| compose via conjunction vs a GAN/LACE/GLIDE baseline | Composable Diffusion · CLEVR (3 components) | FID: 10.51 vs 19.01 (StyleGAN2) | Improved |
| compose three relations with conjunction vs EBM | Composable Diffusion · Relational CLEVR (3 components) | FID: 26.11 vs 58.66 (EBM) | Improved |
| compose relations with conjunction vs EBM | Composable Diffusion · Relational CLEVR (1 component) | classification accuracy (%): 60.40 vs 78.14 (EBM) | Hurt |
| compose relations with conjunction vs GLIDE | Composable Diffusion · Relational CLEVR (2 components) | FID: 29.82 vs 28.56 (GLIDE) | Mixed |
| compose facial attributes with conjunction/negation vs LACE | Composable Diffusion · FFHQ (3 components) | classification accuracy (%): 68.86 vs 80.88 (LACE) | Hurt |
| compose facial attributes vs EBM | Composable Diffusion · FFHQ (3 components) | FID: 16.95 vs 335.70 (EBM) | Improved |
| compose three facial attributes vs GLIDE | Composable Diffusion · FFHQ (3 components) | classification accuracy (%): 68.86 vs 27.24 (GLIDE) | Improved |
| use compositional operators on multiple prompts instead of a single long sentence prompt | GLIDE · text-to-image qualitative | qualitative | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- GLIDE — GLIDE: Towards Photorealistic Image Generation and Editing with Text-Guided Diffusion Models (ICML 2022) · better on 6 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
- Compositional Discrete Latent Code for High Fidelity, Productive Diffusion Models (NeurIPS 2025 · similarity 0.89)
- RealCompo: Balancing Realism and Compositionality Improves Text-to-Image Diffusion Models (NeurIPS 2024 · similarity 0.88)
- Compositional Scene Understanding through Inverse Generative Modeling (ICML 2025 · similarity 0.88)
- Generating compositional scenes via Text-to-image RGBA Instance Generation (NeurIPS 2024 · similarity 0.88)
- Compositional Image Decomposition with Diffusion Models (ICML 2024 · similarity 0.88)
- Mastering Text-to-Image Diffusion: Recaptioning, Planning, and Generating with Multimodal LLMs (ICML 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.
