model_zoo.md
March 16, 2026 · View on GitHub
🔥 1. We provide all the links of Sana pth and diffusers safetensor below
SANA
SANA-1.5
| Model | Reso | pth link | diffusers | Precision | Description |
|---|---|---|---|---|---|
| SANA1.5-4.8B | 1024px | SANA1.5_4.8B_1024px | Efficient-Large-Model/SANA1.5_4.8B_1024px_diffusers | bf16 | Multi-Language |
| SANA1.5-1.6B | 1024px | SANA1.5_1.6B_1024px | Efficient-Large-Model/SANA1.5_1.6B_1024px_diffusers | bf16 | Multi-Language |
SANA-Sprint
| Model | Reso | pth link | diffusers | Precision | Description |
|---|---|---|---|---|---|
| Sana-Sprint-0.6B | 1024px | Sana-Sprint_0.6B_1024px | Efficient-Large-Model/Sana_Sprint_0.6B_1024px_diffusers | bf16 | Multi-Language |
| Sana-Sprint-1.6B | 1024px | Sana-Sprint_1.6B_1024px | Efficient-Large-Model/Sana_Sprint_1.6B_1024px_diffusers | bf16 | Multi-Language |
SANA-Video
| Model | Reso | pth link | diffusers | Precision | Description |
|---|---|---|---|---|---|
| Sana-Video-2B | 480p | Sana-Video_2B_480p | Efficient-Large-Model/Sana-Video_2B_480p_diffusers | bf16 | 5s Pre-train model |
| Sana-Video-2B | 720p | Sana-Video_2B_720p | Efficient-Large-Model/SANA-Video_2B_720p_diffusers | bf16 | 5s 720p model (LTX2 VAE) |
| LongSANA-Video-2B | 480p | SANA-Video_2B_480p_LongLive | Efficient-Large-Model/SANA-Video_2B_480p_LongLive_diffusers | bf16 | 27FPS Minute-length model |
| LongSANA-Video-2B-ODE-Init | 480p | LongSANA_2B_480p_ode | --- | bf16 | LongSANA first step model initialized from ODE trajectories |
| LongSANA-Video-2B-Self-Forcing | 480p | LongSANA_2B_480p_self_forcing | --- | bf16 | LongSANA second step model trained by Self-Forcing |
❗ 2. Make sure to use correct precision(fp16/bf16/fp32) for training and inference.
We provide two samples to use fp16 and bf16 weights, respectively.
❗️Make sure to set variant and torch_dtype in diffusers pipelines to the desired precision.
1). For fp16 models
# run `pip install git+https://github.com/huggingface/diffusers` before use Sana in diffusers
import torch
from diffusers import SanaPipeline
pipe = SanaPipeline.from_pretrained(
"Efficient-Large-Model/Sana_1600M_1024px_diffusers",
variant="fp16",
torch_dtype=torch.float16,
)
pipe.to("cuda")
pipe.vae.to(torch.bfloat16)
pipe.text_encoder.to(torch.bfloat16)
prompt = 'a cyberpunk cat with a neon sign that says "Sana"'
image = pipe(
prompt=prompt,
height=1024,
width=1024,
guidance_scale=5.0,
num_inference_steps=20,
generator=torch.Generator(device="cuda").manual_seed(42),
)[0]
image[0].save("sana.png")
2). For bf16 models
# run `pip install git+https://github.com/huggingface/diffusers` before use Sana in diffusers
import torch
from diffusers import SanaPAGPipeline
pipe = SanaPAGPipeline.from_pretrained(
"Efficient-Large-Model/Sana_1600M_1024px_BF16_diffusers",
variant="bf16",
torch_dtype=torch.bfloat16,
pag_applied_layers="transformer_blocks.8",
)
pipe.to("cuda")
pipe.text_encoder.to(torch.bfloat16)
pipe.vae.to(torch.bfloat16)
prompt = 'a cyberpunk cat with a neon sign that says "Sana"'
image = pipe(
prompt=prompt,
guidance_scale=5.0,
pag_scale=2.0,
num_inference_steps=20,
generator=torch.Generator(device="cuda").manual_seed(42),
)[0]
image[0].save('sana.png')
❗ 3. 2K & 4K models
4K models need VAE tiling to avoid OOM issue.(16 GPU is recommended)
# run `pip install git+https://github.com/huggingface/diffusers` before use Sana in diffusers
import torch
from diffusers import SanaPipeline
# 2K model: Efficient-Large-Model/Sana_1600M_2Kpx_BF16_diffusers
# 4K model:Efficient-Large-Model/Sana_1600M_4Kpx_BF16_diffusers
pipe = SanaPipeline.from_pretrained(
"Efficient-Large-Model/Sana_1600M_4Kpx_BF16_diffusers",
variant="bf16",
torch_dtype=torch.bfloat16,
)
pipe.to("cuda")
pipe.vae.to(torch.bfloat16)
pipe.text_encoder.to(torch.bfloat16)
# for 4096x4096 image generation OOM issue, feel free adjust the tile size
if pipe.transformer.config.sample_size == 128:
pipe.vae.enable_tiling(
tile_sample_min_height=1024,
tile_sample_min_width=1024,
tile_sample_stride_height=896,
tile_sample_stride_width=896,
)
prompt = 'a cyberpunk cat with a neon sign that says "Sana"'
image = pipe(
prompt=prompt,
height=4096,
width=4096,
guidance_scale=5.0,
num_inference_steps=20,
generator=torch.Generator(device="cuda").manual_seed(42),
)[0]
image[0].save("sana_4K.png")
❗ 4. int4 inference
This int4 model is quantized with SVDQuant-Nunchaku. You need first follow the guidance of installation of nunchaku engine, then you can use the following code snippet to perform inference with int4 Sana model.
Here we show the code snippet for SanaPipeline. For SanaPAGPipeline, please refer to the SanaPAGPipeline section.
import torch
from diffusers import SanaPipeline
from nunchaku.models.transformer_sana import NunchakuSanaTransformer2DModel
transformer = NunchakuSanaTransformer2DModel.from_pretrained("mit-han-lab/svdq-int4-sana-1600m")
pipe = SanaPipeline.from_pretrained(
"Efficient-Large-Model/Sana_1600M_1024px_BF16_diffusers",
transformer=transformer,
variant="bf16",
torch_dtype=torch.bfloat16,
).to("cuda")
pipe.text_encoder.to(torch.bfloat16)
pipe.vae.to(torch.bfloat16)
image = pipe(
prompt="A cute 🐼 eating 🎋, ink drawing style",
height=1024,
width=1024,
guidance_scale=4.5,
num_inference_steps=20,
generator=torch.Generator().manual_seed(42),
).images[0]
image.save("sana_1600m.png")
🔧 5. Convert .pth to diffusers .safetensor
python tools/convert_scripts/convert_sana_to_diffusers.py \
--orig_ckpt_path Efficient-Large-Model/Sana_1600M_1024px_BF16/checkpoints/Sana_1600M_1024px_BF16.pth \
--model_type SanaMS_1600M_P1_D20 \
--dtype bf16 \
--dump_path output/Sana_1600M_1024px_BF16_diffusers \
--save_full_pipeline