What Is Sana?
Sana is developed by NVIDIA. NVIDIA’s efficient high-resolution image model with fast generation, strong alignment and flexible styles. This page maps only to the named model contract, so changing inference infrastructure never changes the model advertised to the user.
What Is Sana Best For?
- Fast high-resolution art
- 4K open-model output
- Style-diverse generation
How to Use Sana
- Describe the subject, composition, light and intended medium.
- Choose a native aspect ratio.
- Adjust only the controls this exact model exposes.
- Generate a draft and inspect details before publishing.
Sana Settings Explained
Aspect Ratio
Only ratios accepted by this exact route appear in the selector.
Credits
The displayed amount includes the selected resolution and any billable quality control.
The panel above shows only the parameters Sana actually supports. Switch versions and unsupported controls disappear rather than sitting there doing nothing.
Sana Prompt Examples
Editorial scene
Create a refined editorial image with coherent light, intentional composition and material detail, composed specifically to evaluate Sana
Product concept
Studio product concept on a clean architectural set, realistic reflections, precise negative space, art direction suited to Sana
Sana Test Results
Testing in progress
Every model runs the same eight-category prompt set — portrait, product, poster text, anime character, interior, multi-reference, image editing and character consistency — recording generation time, success rate, text accuracy and identity retention. That run has not finished for Sana, so this page shows no scores. We do not publish numbers we have not measured.
Sana Limitations
- This page exposes the verified generation contract only
- Model-specific safety and license terms still apply
- Shared benchmark results are still being collected
Sana Pricing and Credits
Credits scale with resolution and image count. The Generate button always shows what this run costs.
1K
1 credits / image
2K
2 credits / image
4K
4 credits / image
Failed generations cost 0 credits.
Commercial Use and Image Rights
Usage is governed by the model’s license and the selected service terms; review both before commercial deployment.
- Model used: Sana
- Model developer: NVIDIA
- The inference route is managed internally and may change without changing the selected model
- Uploads are sent only to the configured inference service needed to complete the generation
- You can delete any generation from your history at any time
Frequently Asked Questions
Who developed Sana?+
NVIDIA is shown as the model developer; inference infrastructure is deliberately kept separate.
Can the route silently substitute another model?+
No. Paid routing requires an exact model match and an explicitly implemented request mapping.
Release History and Sources
November 2024 · 1.6B weights and code released
November 2024 refers to the 1.6B weights and code. The research paper and demo appeared earlier; they are separate events.
Explore this event on the image model timeline →