Instructions to use SceneWorks/Sana_1600M_1024px_mlx with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use SceneWorks/Sana_1600M_1024px_mlx with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] hf download SceneWorks/Sana_1600M_1024px_mlx --local-dir Sana_1600M_1024px_mlx
- Sana
How to use SceneWorks/Sana_1600M_1024px_mlx with Sana:
# Load the model and infer image from text import torch from app.sana_pipeline import SanaPipeline from torchvision.utils import save_image sana = SanaPipeline("configs/sana_config/1024ms/Sana_1600M_img1024.yaml") sana.from_pretrained("hf://SceneWorks/Sana_1600M_1024px_mlx") image = sana( prompt='a cyberpunk cat with a neon sign that says "Sana"', height=1024, width=1024, guidance_scale=5.0, pag_guidance_scale=2.0, num_inference_steps=18, ) - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Atomic Chat
Commit History
23 days ago
-
Relicense to Apache-2.0 (NVIDIA 2026-07-31) + Gemma Terms of Use for the text encoder
Verified signaturebdd56db
-
fix(sana_1600m): re-host transformer from Sana_1600M_1024px_BF16 (sc-11760)
Verified signatureReplace q4/q8/bf16 transformer weights (were converted from the bad Sana_1600M_1024px checkpoint -> petal-blob noise, diffusers #10241) with the good Sana_1600M_1024px_BF16 checkpoint. VAE (DC-AE) + Gemma-2 TE unchanged. Verified clean on the native MLX port (flow-match Euler) across q4/q8/bf16.ac42169
-
Add bf16 quant tier (sc-8489, epic 8506): packed Linear-DiT + Gemma-2 TE, dense DC-AE VAE
Verified signatureba22f36 -
Add q8 quant tier (sc-8489, epic 8506): packed Linear-DiT + Gemma-2 TE, dense DC-AE VAE
Verified signatured7348ea -
Add q4 quant tier (sc-8489, epic 8506): packed Linear-DiT + Gemma-2 TE, dense DC-AE VAE
Verified signaturee68dac8
-
Add SANA MLX mirror (verbatim repackage of NVIDIA diffusers checkpoint into SanaPipeline::from_snapshot layout)
Verified signatureeef346e -
initial commit
Verified signature371b04e