AI Paper Atlas

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

NeurIPS 2023 · Vision-language and multimodal › Human and embodied modeling · approach: Vision-language pretraining · 6 citations (OpenAlex)

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 changedWhereMeasuredEffect
use multimodal prompt (caption plus reference image) instead of textMultiFusion · MS-COCO val, guidance 4.0FID-30k: 9.90 -> 8.53Improved
use image-only prompt instead of textMultiFusion · MS-COCO val, guidance 4.0FID-30k: 9.90 -> 6.03Improved
MultiFusion text prompt vs SD v1.5MultiFusion · MS-COCO val, guidance 4.0FID-30k: 10.19 -> 9.90No clear effect
MultiFusion text prompt vs SD v1.5 at low guidanceMultiFusion · MS-COCO val, guidance 2.0FID-30k: 9.74 -> 12.21Hurt
MultiFusion vs SD v1.5 text-image alignmentMultiFusion · MS-COCO val, guidance 4.0CLIPScore: 0.31 -> 0.29Hurt
multimodal vs text prompts on compositional generationMultiFusion · MCC-250two objects with correct colors (%): 21.66 -> 58.35Improved
MultiFusion multimodal vs Stable DiffusionMultiFusion · MCC-250two objects with correct colors (%): 29.92 -> 58.35Improved
Composable Diffusion vs Stable DiffusionComposable Diffusion · MCC-250two objects with correct colors (%): 29.92 -> 25.59Hurt
MultiFusion multimodal vs text on animal-only subsetMultiFusion · MCC-250 animalstwo objects with correct colors (%): 21.38 -> 21.38No clear effect
MultiFusion vs SD on animal-only compositionsMultiFusion · MCC-250 animalstwo objects with correct colors (%): 31.99 -> 21.38Hurt
increase text weight in attention manipulation 1:1 to 20:1 (ablation)MultiFusion multimodal · MS-COCO val, guidance 4.0FID-30k: 6.74 -> 9.22Hurt
text weight 10:1 vs 1:1 (ablation)MultiFusion multimodal · MS-COCO val, guidance 4.0FID-30k: 6.74 -> 8.53Hurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy