This model requires you to accept the NVIDIA Open Model license terms before download.
Description
SANA 1600M 4K — NVIDIA Labs efficient text-to-image diffusion model that generates ultra-high-resolution images up to 4096×4096px using a deep-compression autoencoder and a linear-attention transformer for fast inference. BF16 weights.
Tags
At a glance
| Kind | image generation model |
|---|---|
| Precision | float32 |
| Container size | 12.0 GiB |
| Weights inside | 12.0 GiB |
| Components | 3 |
| Licence | NVIDIA Open Model |
| Origin | Efficient-Large-Model/Sana_1600M_4Kpx_BF16_diffusers @ 0f1d3414c1c4 |
What is inside the container
A .nbx container packs every part of Sana_1600M_4Kpx_BF16_diffusers — 3 components with weights, graphs and runtime settings — so one file is all the runtime needs.
| Component | Blocks | Layers | Weights |
|---|---|---|---|
| transformer | 20 | 20 | 6.0 GiB |
| vae | 6 | — | 1.2 GiB |
| text_encoder | 26 | — | 4.9 GiB |
Default settings
The values the container ships with. They apply when you do not pass your own.
- Height (px)
- 4096
- Width (px)
- 4096
- Inference steps
- 20
- Guidance scale
- 4.5
Memory needed
The weights of Sana_1600M_4Kpx_BF16_diffusers take 12.0 GiB in float32. That is the floor for the memory the runtime has to place, spread over the devices it finds; activations come on top and grow with the size and length you ask for. A measured peak per device class is not published for this model yet.
Verify what you downloaded
Every download carries its SHA-256, and the file endpoint sends it in the X-NBX-SHA256 header. The container is 12.0 GiB (12,903,903,319 bytes).
$ sha256sum Sana_1600M_4Kpx_BF16_diffusers.nbx
923b7d39a19c20ecdec0a96222908270a9b77e0f0cfb996b432d9d9ccf276216
Questions
What is Sana_1600M_4Kpx_BF16_diffusers?
A image generation model by Efficient-Large-Model, distributed as a float32 .nbx container and licensed NVIDIA Open Model.
How big is the download?
12.0 GiB, of which 12.0 GiB are weights.
How do I run it?
Install the runtime from the download page, then use the commands under “CLI Usage” below: neurobrix import Efficient-Large-Model/Sana_1600M_4Kpx_BF16_diffusers, then neurobrix serve --model Sana_1600M_4Kpx_BF16_diffusers.
CLI Usage
$ neurobrix import Efficient-Large-Model/Sana_1600M_4Kpx_BF16_diffusers
$ neurobrix serve --model Sana_1600M_4Kpx_BF16_diffusers
$ neurobrix run --prompt "your prompt here"
$ neurobrix stop