Nvidia Cosmos 3 Super Text-to-Image API
Cosmos3 is a collection of Omnimodal world models capable of generating dynamic, high-quality video, image, audio, and action commands from combinations of text, image, video, and action trajectory inputs.
NVIDIA Cosmos 3 Super Text-to-Image
NVIDIA Cosmos 3 Super Text-to-Image is NVIDIA's flagship text-to-image model built on the Cosmos 3 architecture. Designed for high-fidelity physical world simulation, it delivers photorealistic images with strong prompt adherence, precise spatial reasoning, and support for negative prompting — making it well suited for both creative and technical generation tasks.
🌟 Key Features
🎨 Photorealistic Generation
Produces high-quality images grounded in physical realism, excelling at scenes with natural lighting, accurate materials, and spatial coherence.
📝 Strong Prompt Fidelity
Understands complex, multi-element prompts and accurately reflects subject, environment, style, and compositional instructions in the output.
🚫 Negative Prompting
Accepts a negative_prompt to explicitly exclude unwanted content, artifacts, or styles from the generated image.
📐 Flexible Output Sizes
Choose from six standard presets covering square, portrait, and landscape orientations to match any platform or use case.
🔁 Reproducible Results
Pin a seed value to reproduce or iterate on a specific generation, enabling systematic prompt refinement workflows.
⚙️ Parameters
| Parameter | Required | Description |
|---|---|---|
prompt | ✅ | Text description of the image to generate. Maximum 2,500 characters. |
negative_prompt | ❌ | Content to exclude from the output (default: empty). |
image_size | ❌ | Output size preset: square_hd, square, portrait_4_3, portrait_16_9, landscape_4_3, landscape_16_9 (default). |
num_inference_steps | ❌ | Denoising steps (1–100, default: 28). Higher values improve quality but increase time. |
guidance_scale | ❌ | Prompt adherence strength (1.0–20.0, default: 4.0). Higher values follow the prompt more strictly. |
output_format | ❌ | Output format: jpeg (default) or png. |
seed | ❌ | Random seed for reproducibility. Omit for random results. |
💲 Pricing
| Billing Standard | Price |
|---|---|
| Per image | $0.044 |
Billing is based on the number of images generated.
💡 Prompt Tips
A photorealistic close-up of hands shaping clay on a pottery wheel, warm studio lighting, shallow depth of field, 8K resolution.
- Describe subject, environment, lighting, and camera perspective in a single coherent prompt.
- Use
negative_promptto remove common artifacts:blurry, low quality, distorted, watermark, cartoon. - Keep
guidance_scalebetween 3–5 for natural-looking outputs; increase to 7–10 for strict prompt adherence. - Use
landscape_16_9for cinematic or widescreen compositions. - Fix
seedwhen iterating on prompt wording to isolate the effect of each change.
🎯 Use Cases
- Creative Illustration — Generate concept art, characters, and environments from descriptive text prompts.
- Product Visualization — Produce photorealistic product renders and lifestyle imagery from written descriptions.
- Scene Prototyping — Rapidly visualize environment and set designs before committing to production.
- Social Media Content — Create scroll-stopping visuals in any aspect ratio for any platform.
- Research & Simulation — Leverage Cosmos's physical world grounding for technically accurate scene generation.
- Rapid Iteration — Generate multiple candidates with varied seeds and prompts to explore creative directions quickly.
📝 Notes
promptis the only required field (besidesmodel).output_formatsupportsjpeg(smaller file size) andpng(lossless).- Task status values:
created,processing,completed,failed. - Generated image URLs are returned in
outputsonce status iscompleted.
🔗 Related Models
- NVIDIA Cosmos 3 Super Image-to-Video — Animate a reference image into a video clip with a text prompt.
