Papers
arxiv:2410.10629

SANA: Efficient High-Resolution Image Synthesis with Linear Diffusion Transformers

Published on Oct 14, 2024
Authors:
,
,
,
,
,

Abstract

We introduce Sana, a text-to-image framework that can efficiently generate images up to 4096times4096 resolution. Sana can synthesize high-resolution, high-quality images with strong text-image alignment at a remarkably fast speed, deployable on laptop GPU. Core designs include: (1) Deep compression autoencoder: unlike traditional AEs, which compress images only 8times, we trained an AE that can compress images 32times, effectively reducing the number of latent tokens. (2) Linear DiT: we replace all vanilla attention in DiT with linear attention, which is more efficient at high resolutions without sacrificing quality. (3) Decoder-only text encoder: we replaced T5 with modern decoder-only small LLM as the text encoder and designed complex human instruction with in-context learning to enhance the image-text alignment. (4) Efficient training and sampling: we propose Flow-DPM-Solver to reduce sampling steps, with efficient caption labeling and selection to accelerate convergence. As a result, Sana-0.6B is very competitive with modern giant diffusion model (e.g. Flux-12B), being 20 times smaller and 100+ times faster in measured throughput. Moreover, Sana-0.6B can be deployed on a 16GB laptop GPU, taking less than 1 second to generate a 1024times1024 resolution image. Sana enables content creation at low cost. Code and model will be publicly released.

Community

Sign up or log in to comment

Models citing this paper 10

Browse 10 models citing this paper

Datasets citing this paper 0

No dataset linking this paper

Cite arxiv.org/abs/2410.10629 in a dataset README.md to link it from this page.

Spaces citing this paper 2

Collections including this paper 5