MUMU: Bootstrapping Multimodal Image Generation from Text-to-Image Data

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Berman, William, Peysakhovich, Alexander
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910600821473280
author Berman, William
Peysakhovich, Alexander
author_facet Berman, William
Peysakhovich, Alexander
contents We train a model to generate images from multimodal prompts of interleaved text and images such as "a <picture of a man> man and his <picture of a dog> dog in an <picture of a cartoon> animated style." We bootstrap a multimodal dataset by extracting semantically meaningful image crops corresponding to words in the image captions of synthetically generated and publicly available text-image data. Our model, MUMU, is composed of a vision-language model encoder with a diffusion decoder and is trained on a single 8xH100 GPU node. Despite being only trained on crops from the same image, MUMU learns to compose inputs from different images into a coherent output. For example, an input of a realistic person and a cartoon will output the same person in the cartoon style, and an input of a standing subject and a scooter will output the subject riding the scooter. As a result, our model generalizes to tasks such as style transfer and character consistency. Our results show the promise of using multimodal models as general purpose controllers for image generation.
format Preprint
id arxiv_https___arxiv_org_abs_2406_18790
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle MUMU: Bootstrapping Multimodal Image Generation from Text-to-Image Data
Berman, William
Peysakhovich, Alexander
Computer Vision and Pattern Recognition
Artificial Intelligence
We train a model to generate images from multimodal prompts of interleaved text and images such as "a <picture of a man> man and his <picture of a dog> dog in an <picture of a cartoon> animated style." We bootstrap a multimodal dataset by extracting semantically meaningful image crops corresponding to words in the image captions of synthetically generated and publicly available text-image data. Our model, MUMU, is composed of a vision-language model encoder with a diffusion decoder and is trained on a single 8xH100 GPU node. Despite being only trained on crops from the same image, MUMU learns to compose inputs from different images into a coherent output. For example, an input of a realistic person and a cartoon will output the same person in the cartoon style, and an input of a standing subject and a scooter will output the subject riding the scooter. As a result, our model generalizes to tasks such as style transfer and character consistency. Our results show the promise of using multimodal models as general purpose controllers for image generation.
title MUMU: Bootstrapping Multimodal Image Generation from Text-to-Image Data
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2406.18790