본문 바로가기
AI/StableDiffusionWebUI

wuerstchen

by wenect 2023. 9. 17.

https://huggingface.co/blog/wuerstchen

 

Introducing Würstchen: Fast Diffusion for Image Generation

Introducing Würstchen: Fast Diffusion for Image Generation What is Würstchen? Würstchen is a diffusion model, whose text-conditional component works in a highly compressed latent space of images. Why is this important? Compressing data can reduce comput

huggingface.co

https://huggingface.co/docs/diffusers/main/en/api/pipelines/wuerstchen

 

Würstchen

Taking Diffusers Beyond Images

huggingface.co

StableDiffusion보다 속도가 2배이상 빠름

Würstchen은 Stable Diffusion과 같은 확산 모델보다 더 빠르게 이미지를 생성하는 동시에 훨씬 적은 메모리를 소비하여 비슷한 결과를 얻는 개방형 텍스트-이미지 모델입니다. 이 접근 방식은 텍스트 조건부 확산 모델, 이미지 인코더/디코더 및 VQGAN의 세 가지 모델 파이프라인을 기반으로 합니다. Stable Diffusion 1.4 대비 모델 학습 비용이 16배 절감되었습니다. 이 성과는 압축 단계를 추가하여 512×512 이미지를 42배 줄여 12×12의 해상도로 달성함으로써 가능해졌습니다. Huggingface 에서 모델 데모를 시도해 볼 수 있으며 오픈 소스 코드는 Github 에서 사용할 수 있습니다 .

https://github.com/dome272/wuerstchen

 

GitHub - dome272/Wuerstchen: Official implementation of Würstchen: Efficient Pretraining of Text-to-Image Models

Official implementation of Würstchen: Efficient Pretraining of Text-to-Image Models - GitHub - dome272/Wuerstchen: Official implementation of Würstchen: Efficient Pretraining of Text-to-Image Models

github.com

SDXL에 비해 두배이상 빠른 속도를 보인다.

'AI > StableDiffusionWebUI' 카테고리의 다른 글

ComfyUI Portrait Master 2.3  (0) 2023.12.24
StableDifusion xl movie  (0) 2023.12.24
prompt Camera angle  (0) 2023.09.03
SDv1.5 vs SDv2.1 vs SDXL  (0) 2023.07.31
Stable Diffusion - QR Code Generation  (3) 2023.06.15

댓글