Research map / Vision-language and multimodal / Human and embodied modeling
MultiFusion: Fusing Pre-Trained Models for Multi-Lingual, Multi-Modal Image Generation
Marco Bellagente, Manuel Brack, Hannah Benita Teufel, Felix Friedrich, Björn Deiseroth, Constantin Eichenberg, Andrew Dai, Robert John Nicholas Baldock, Souradeep Nanda, Koen Oostermeijer, Andres Felipe Cruz-Salinas, Patrick Schramowski and 2 more
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 |
|---|---|---|---|
| use multimodal prompt (caption plus reference image) instead of text | MultiFusion · MS-COCO val, guidance 4.0 | FID-30k: 9.90 -> 8.53 | Improved |
| use image-only prompt instead of text | MultiFusion · MS-COCO val, guidance 4.0 | FID-30k: 9.90 -> 6.03 | Improved |
| MultiFusion text prompt vs SD v1.5 | MultiFusion · MS-COCO val, guidance 4.0 | FID-30k: 10.19 -> 9.90 | No clear effect |
| MultiFusion text prompt vs SD v1.5 at low guidance | MultiFusion · MS-COCO val, guidance 2.0 | FID-30k: 9.74 -> 12.21 | Hurt |
| MultiFusion vs SD v1.5 text-image alignment | MultiFusion · MS-COCO val, guidance 4.0 | CLIPScore: 0.31 -> 0.29 | Hurt |
| multimodal vs text prompts on compositional generation | MultiFusion · MCC-250 | two objects with correct colors (%): 21.66 -> 58.35 | Improved |
| MultiFusion multimodal vs Stable Diffusion | MultiFusion · MCC-250 | two objects with correct colors (%): 29.92 -> 58.35 | Improved |
| Composable Diffusion vs Stable Diffusion | Composable Diffusion · MCC-250 | two objects with correct colors (%): 29.92 -> 25.59 | Hurt |
| MultiFusion multimodal vs text on animal-only subset | MultiFusion · MCC-250 animals | two objects with correct colors (%): 21.38 -> 21.38 | No clear effect |
| MultiFusion vs SD on animal-only compositions | MultiFusion · MCC-250 animals | two objects with correct colors (%): 31.99 -> 21.38 | Hurt |
| increase text weight in attention manipulation 1:1 to 20:1 (ablation) | MultiFusion multimodal · MS-COCO val, guidance 4.0 | FID-30k: 6.74 -> 9.22 | Hurt |
| text weight 10:1 vs 1:1 (ablation) | MultiFusion multimodal · MS-COCO val, guidance 4.0 | FID-30k: 6.74 -> 8.53 | Hurt |
Closest papers by meaning
- Exploring the Deep Fusion of Large Language Models and Diffusion Transformers for Text-to-Image Synthesis (CVPR 2025 · similarity 0.92)
- LMFusion: Adapting Pretrained Language Models for Multimodal Generation (NeurIPS 2025 · similarity 0.91)
- MuLan: Adapting Multilingual Diffusion Models for Hundreds of Languages with Negligible Cost (ICML 2025 · similarity 0.91)
- Dual Diffusion for Unified Image Generation and Understanding (CVPR 2025 · similarity 0.91)
- AltDiffusion: A Multilingual Text-to-Image Diffusion Model (AAAI 2024 · similarity 0.90)
- Instruct-Imagen: Image Generation with Multi-modal Instruction (CVPR 2024 · 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.
