Describe the picture in your own words: what's in it, the style (photo, cartoon, painting…), the colors and the mood. More detail gets you closer to what you imagine.

Aspect Ratio

The shape of the result. 16:9 is wide, like YouTube or a TV. 9:16 is tall, like TikTok or phone stories. 1:1 is square, like an Instagram post.

Sana 1
Sana 2
Sana 3
Sana 4
Sana 5
Sana 6
Sana 7
Sana 8
Sana 9

NVIDIA · Text to image

Sana AI Image Generator

Sana is NVIDIA's small, very efficient image model, built for speed, with a negative prompt and a seed. Use Sana to compare styles for the same subject quickly and at low cost.

Real results

Discover What Sana Can Create

Posters, watercolours, storybook art and wallpapers, each made from one prompt with Sana. Unedited outputs from this page, with the prompt for each.

Loose watercolour landscape of rolling hills with a red barn, wet-on-wet sky, white paper showing through

Watercolour landscape · Sana

Loose watercolour landscape of rolling hills with a red barn, wet-on-wet sky, white paper showing through

Key Features of Sana

Sana comes from NVIDIA researchers working with MIT, published in late 2024. It is a linear-attention diffusion transformer that generates high-resolution images with far fewer parameters and far less compute than models like FLUX, small enough to run on a laptop GPU. ImageFX runs Sana 1.6B, the larger of the two released sizes, at 1024-pixel resolution.

01 / 04

Efficiency by Design

Sana generates at high resolution with a fraction of the parameters of larger models, which keeps it fast and cheap to run. That makes Sana a good sandbox: try a subject in five styles before you commit to a bigger model.

Try Sana Now
Watercolour landscape
Prompt · Loose watercolour landscape of rolling hills with a red barn, wet-on-wet sky, white paper showing through

02 / 04

Wide Artistic Range

NVIDIA highlights Sana's range of styles: painterly, graphic, photographic and fantasy prompts all land reasonably well. Lead with an explicit style word such as "watercolour", "neon cyberpunk poster" or "pencil sketch", and keep the scene itself simple.

Explore Styles with Sana
Cyberpunk poster
Prompt · Neon cyberpunk poster of a cat wearing a visor, magenta and cyan glow, rain streaks, bold graphic shapes. Negative: photo, blurry

03 / 04

Negative Prompt and Seed

Sana takes both a negative prompt and a seed on ImageFX, so you can push away unwanted elements and repeat results. Use the negative prompt for the one or two things that keep going wrong, such as text or people.

Use a Negative Prompt with Sana
Abstract wallpaper
Prompt · 16:9 abstract wallpaper of flowing liquid gold and deep navy, smooth gradients, minimal. Negative: text, people

04 / 04

Illustration for Stories and Books

Soft, stylised illustration is where Sana is most at home. Describe one character, one action and the texture you want, and Sana keeps the look consistent with the style words.

Make an Illustration with Sana
Storybook illustration
Prompt · Gentle storybook illustration of a hedgehog carrying a lantern through a snowy forest, soft pencil texture, warm light

How to Use Sana on ImageFX

  1. Lead with the Style

    Open the Sana generator at the top of this page. Give it a clear subject and an explicit style, and keep the scene simple.

  2. Add a Negative Prompt and Pick the Shape

    Under More Options, list what you don't want in the negative prompt and set a seed if you want to repeat a result. Then choose an aspect ratio.

  3. Generate and Download

    Click Generate, compare a few runs, then download the PNG. Every image is saved to your account.

A Sana run costs 4 credits; the Generate button shows the price first, and a failed run costs nothing. See plans and credit packs

FAQs About Sana

Who made Sana?

Researchers at NVIDIA, working with MIT. Sana was published in late 2024.

Which Sana version runs on ImageFX?

Sana 1.6B, the 1.6-billion-parameter model, at 1024-pixel resolution.

Does Sana support negative prompts?

Yes. The negative prompt field is passed to Sana, and you can also set a seed.

What size are Sana images?

Each aspect ratio maps to a set size around one megapixel, for example 1024 × 1024 for a square image.

Is Sana good for photorealism?

Sana can do it, but Photon or FLUX models are the better choice for realistic photos.

Can Sana put text in images?

Not reliably. Treat any lettering from Sana as decorative, and use Ideogram or Recraft when the words matter.

What are Sana's limits?

At 1.6 billion parameters, fine detail and complex scenes are not Sana's strength. Several subjects with specific positions can get mixed up; simplify the prompt, or use a FLUX model.

How does Sana compare with Flux Schnell and SD 3.5 Large?

Flux Schnell costs less and is stronger on photographic prompts; Sana adds a negative prompt and is the better stylist. SD 3.5 Large is a much bigger model with more detail; Sana is lighter and quicker for style tests.

Try Sana on ImageFX

Test styles and ideas quickly with NVIDIA's efficient image model, with a negative prompt and a seed when you need them.

Create with Sana