⚡️Sana: Efficient High-Resolution Image Synthesis with Linear Diffusion Transformer

         

💡 Introduction

We introduce Sana, a text-to-image framework that can efficiently generate images up to 4096 × 4096 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) DC-AE: unlike traditional AEs, which compress images only 8×, we trained an AE that can compress images 32×, 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 1024 × 1024 resolution image. Sana enables content creation at low cost.

🔥🔥 News

• (🔥 New) [2024/11] 1.6B Sana models are released.
• (🔥 New) [2024/11] Training & Inference & Metrics code are released.
• (🔥 New) [2024/11] Working on diffusers.
• [2024/10] Demo is released.
• [2024/10] DC-AE Code and weights are released!
• [2024/10] Paper is on Arxiv!

Performance

Methods (1024x1024)	Throughput (samples/s)	Latency (s)	Params (B)	Speedup	FID 👆	CLIP 👆	GenEval 👆	DPG 👆
FLUX-dev	0.04	23.0	12.0	1.0×	10.15	27.47	0.67	84.0
Sana-0.6B	1.7	0.9	0.6	39.5×	5.81	28.36	0.64	83.6
Sana-1.6B	1.0	1.2	1.6	23.3×	5.76	28.67	0.66	84.8

Click to show all

Methods	Throughput (samples/s)	Latency (s)	Params (B)	Speedup	FID 👆	CLIP 👆	GenEval 👆	DPG 👆
512 × 512 resolution
PixArt-α	1.5	1.2	0.6	1.0×	6.14	27.55	0.48	71.6
PixArt-Σ	1.5	1.2	0.6	1.0×	6.34	27.62	0.52	79.5
Sana-0.6B	6.7	0.8	0.6	5.0×	5.67	27.92	0.64	84.3
Sana-1.6B	3.8	0.6	1.6	2.5×	5.16	28.19	0.66	85.5
1024 × 1024 resolution
LUMINA-Next	0.12	9.1	2.0	2.8×	7.58	26.84	0.46	74.6
SDXL	0.15	6.5	2.6	3.5×	6.63	29.03	0.55	74.7
PlayGroundv2.5	0.21	5.3	2.6	4.9×	6.09	29.13	0.56	75.5
Hunyuan-DiT	0.05	18.2	1.5	1.2×	6.54	28.19	0.63	78.9
PixArt-Σ	0.4	2.7	0.6	9.3×	6.15	28.26	0.54	80.5
DALLE3	-	-	-	-	-	-	0.67	83.5
SD3-medium	0.28	4.4	2.0	6.5×	11.92	27.83	0.62	84.1
FLUX-dev	0.04	23.0	12.0	1.0×	10.15	27.47	0.67	84.0
FLUX-schnell	0.5	2.1	12.0	11.6×	7.94	28.14	0.71	84.8
Sana-0.6B	1.7	0.9	0.6	39.5×	5.81	28.36	0.64	83.6
Sana-1.6B	1.0	1.2	1.6	23.3×	5.76	28.67	0.66	84.8

Contents

• Env
• Demo
• Training
• Testing
• TODO
• Citation

🔧 1. Dependencies and Installation

• Python >= 3.10.0 (Recommend to use Anaconda or Miniconda)
• PyTorch >= 2.0.1+cu12.1

git clone https://github.com/NVlabs/Sana.git
cd Sana

./environment_setup.sh sana
# or you can install each components step by step following environment_setup.sh

💻 2. How to Play with Sana (Inference)

💰Hardware requirement

• 9GB VRAM is required for 0.6B model and 12GB VRAM for 1.6B model. Our later quantization version will require less than 8GB for inference.
• All the tests are done on A100 GPUs. Different GPU version may be different.

🔛 Quick start with Gradio

# official online demo
DEMO_PORT=15432 \
python app/app_sana.py \
      --config=configs/sana_config/1024ms/Sana_1600M_img1024.yaml \
      --model_path=hf://Efficient-Large-Model/Sana_1600M_1024px/checkpoints/Sana_1600M_1024px.pth

import torch
from app.sana_pipeline import SanaPipeline
from torchvision.utils import save_image

device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
generator = torch.Generator(device=device).manual_seed(42)

sana = SanaPipeline("configs/sana_config/1024ms/Sana_1600M_img1024.yaml")
sana.from_pretrained("hf://Efficient-Large-Model/Sana_1600M_1024px/checkpoints/Sana_1600M_1024px.pth")
prompt = 'a cyberpunk cat with a neon sign that says "Sana"'

image = sana(
    prompt=prompt,
    height=1024,
    width=1024,
    guidance_scale=5.0,
    pag_guidance_scale=2.0,
    num_inference_steps=18,
    generator=generator,
)
save_image(image, 'output/sana.png', nrow=1, normalize=True, value_range=(-1, 1))

🔛 Run inference with TXT or JSON files

# Run samples in a txt file
python scripts/inference.py \
      --config=configs/sana_config/1024ms/Sana_1600M_img1024.yaml \
      --model_path=hf://Efficient-Large-Model/Sana_1600M_1024px/checkpoints/Sana_1600M_1024px.pth
      --txt_file=asset/samples_mini.txt

# Run samples in a json file
python scripts/inference.py \
      --config=configs/sana_config/1024ms/Sana_1600M_img1024.yaml \
      --model_path=hf://Efficient-Large-Model/Sana_1600M_1024px/checkpoints/Sana_1600M_1024px.pth
      --json_file=asset/samples_mini.json

where each line of asset/samples_mini.txt contains a prompt to generate

🔥 3. How to Train Sana

💰Hardware requirement

• 32GB VRAM is required for both 0.6B and 1.6B model's training

We provide a training example here and you can also select your desired config file from config files dir based on your data structure.

To launch Sana training, you will first need to prepare data in the following formats

asset/example_data
├── AAA.txt
├── AAA.png
├── BCC.txt
├── BCC.png
├── ......
├── CCC.txt
└── CCC.png

Then Sana's training can be launched via

# Example of training Sana 0.6B with 512x512 resolution
bash train_scripts/train.sh \
  configs/sana_config/512ms/Sana_600M_img512.yaml \
  --data.data_dir="[asset/example_data]" \
  --data.type=SanaImgDataset \
  --model.multi_scale=false \
  --train.train_batch_size=32

# Example of training Sana 1.6B with 1024x1024 resolution
bash train_scripts/train.sh \
  configs/sana_config/1024ms/Sana_1600M_img1024.yaml \
  --data.data_dir="[asset/example_data]" \
  --data.type=SanaImgDataset \
  --model.multi_scale=false \
  --train.train_batch_size=8

💻 4. Metric toolkit

Refer to Toolkit Manual.

💪To-Do List

We will try our best to release

• Training code
• Inference code
• [+] Model zoo
• working on Diffusers(huggingface/diffusers#9982)
• ComfyUI
• Laptop development

🤗Acknowledgements

• Thanks to PixArt-α, PixArt-Σ and Efficient-ViT for their wonderful work and codebase!

📖BibTeX

@misc{xie2024sana,
      title={Sana: Efficient High-Resolution Image Synthesis with Linear Diffusion Transformer},
      author={Enze Xie and Junsong Chen and Junyu Chen and Han Cai and Haotian Tang and Yujun Lin and Zhekai Zhang and Muyang Li and Ligeng Zhu and Yao Lu and Song Han},
      year={2024},
      eprint={2410.10629},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2410.10629},
    }