How to run Sana 1.6B (1024px) locally

NVIDIA · Text → image · 1.6B params · NVIDIA Sana Non-Commercial (NSCL) (non-commercial)

NVIDIA's linear-attention DiT with 32x latent compression that generates 1024px (up to 4K) images very fast on low VRAM, licensed for non-commercial use.

What it costs to run — $0 markup

On your own machine — $0. Runs free if you have about 8.0 GB of VRAM (~8GB; extremely fast via deep latent compression (DC-AE 32x)), via Diffusers / ComfyUI.
Rent a GPU — from $0.26/hr. Fits RTX 3090 24GB at the direct vendor price ($0 markup) — pay only for the minutes you generate.
Download the weights — free. Open weights at Efficient-Large-Model/Sana_1600M_1024px.

Note: generative-media models are billed per image / per second / per minute on hosted services — not per token. Running locally or on your own rented GPU is usually far cheaper and keeps your data on your machine.

Key facts

Does	Text → image
VRAM to run	~8.0 GB (~8GB; extremely fast via deep latent compression (DC-AE 32x))
Download	~7 GB
Parameters	1.6B
License	NVIDIA Sana Non-Commercial (NSCL) (non-commercial)
Run with	Diffusers / ComfyUI

Get Sana 1.6B (1024px) on Hugging Face →

More image models

FLUX.1 [dev] — Text → image, ~12 GB VRAM
Stable Diffusion XL 1.0 (SDXL) — Text → image, ~8.0 GB VRAM
Stable Diffusion 1.5 — Text → image, ~4.0 GB VRAM
FLUX.1 Kontext [dev] — Image editing, ~12 GB VRAM
FLUX.1 [schnell] — Text → image, ~12 GB VRAM
Qwen-Image — Text → image, ~24 GB VRAM

Browse: all media models · chat / LLM models