AllImageAI
By NVIDIAActive

NVIDIA Cosmos 3 Super Image Generator

NVIDIA’s high-capacity physical-world model for grounded scenes, materials and cinematic environments.

Ask AI for better images

Generate better images

Prompt ideas, settings and fixes for this model.

If the prompt isn't filled in, copy and paste it into your chat.

Prompt examples Verified
Original AI-generated editorial artwork for Cosmos 3 Super Image
AI-generated showcase artwork for Cosmos 3 Super Image. Editorial cover, not a benchmark result.
Editorial cover

Model Version

Output

Cosmos 3 Super Image

NVIDIA’s high-capacity physical-world model for grounded scenes, materials and cinematic environments.

SlowUp to 1K1:116:99:164:3

Try one of these prompts

Model showcase

Curated outputs with their exact prompts and model provenance.

Developer
NVIDIA
Model
Cosmos 3 Super Image
Status
Active
Max Output
1K
Input
Text
Release date
Not yet verified
Last Verified
August 25, 2026
View release history and date evidence ↗

What Is Cosmos 3 Super Image?

Cosmos 3 Super Text2Image is NVIDIA’s 64B image branch of its omnimodal physical-world family. It is suited to scenes where lighting, material interaction, machinery and spatial relationships matter more than rapid iteration.

What Is Cosmos 3 Super Image Best For?

  • Physical-world scenes
  • Machinery and engineering concepts
  • Cinematic environments
  • Material-rich product visuals

How to Use Cosmos 3 Super Image

  1. Describe objects, materials and their physical relationships.
  2. Keep paid expansion off by writing a complete prompt.
  3. Generate one or two candidates.
  4. Upscale the selected image separately.

Cosmos 3 Super Image Settings Explained

Steps

28 is the documented default and the predictable cost-quality point.

Negative Prompt

Exclude visual artefacts without invoking a paid prompt-rewrite service.

The panel above shows only the parameters Cosmos 3 Super Image actually supports. Switch versions and unsupported controls disappear rather than sitting there doing nothing.

Cosmos 3 Super Image Prompt Examples

Engineering scene

Macro product photograph of a mechanical watch assembling itself, polished steel gears, physically plausible reflections, clean white studio

Grounded landscape

Weathered fishing boat in a misty harbour at dawn, wet timber, soft directional light, realistic water and rope tension

Cosmos 3 Super Image 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 Cosmos 3 Super Image, so this page shows no scores. We do not publish numbers we have not measured.

Cosmos 3 Super Image Limitations

  • 1K output on this integration
  • Slower than compact draft models
  • Agentic generation and paid prompt expansion are intentionally disabled

Cosmos 3 Super Image Pricing and Credits

Credits scale with resolution and image count. The Generate button always shows what this run costs.

1K

3 credits / image

Failed generations cost 0 credits.

Commercial Use and Image Rights

Commercial use is subject to NVIDIA’s Cosmos model licence and the selected service terms.

  • Model used: Cosmos 3 Super Image
  • 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

Why is agentic generation disabled?+

It generates and judges multiple hidden candidates, so cost is not predictable from the requested output count.

Is this the video model?+

No. This page calls the exact Cosmos 3 Super text-to-image operation.

Release History and Sources

Date not yet established

The exact model, checkpoint or provider-specific tier needs to be established before it can be placed on the timeline.

Explore this event on the image model timeline →