# Cosmos 3 Super Image to Video

> Cosmos 3 Super Image to Video — NVIDIA's 64B open-weights image-to-video model (Cosmos3-Super-Image2Video) for physics-real motion, native ambient audio, and browser-based generation. No GPU required.

## About

**Cosmos 3 Super Image to Video** — officially **Cosmos3-Super-Image2Video** — is a post-trained specialist model derived from NVIDIA's **Cosmos 3 Super** foundation model. Released **May 31, 2026** at GTC Taipei / Computex 2026, it animates a single reference image into a temporally coherent, physics-real video clip using a natural-language motion prompt.

The architecture is a **Mixture-of-Transformers (MoT)**: a **32B Reasoner** tower analyzes material properties, gravity, and object relationships before a **32B Generator** diffusion tower renders output. The combined model is **64B parameters**, released as **open weights** on Hugging Face. It ranks **#1 on the open Physics-IQ benchmark** among open-weights models (also cited on PAI-Bench and R-Bench).

**This site supports image-to-video only** — upload one still image plus a motion prompt. There is no standalone text-to-video mode on cosmos3ai.net.

**cosmos3ai.net** is an independent online-generation resource for Cosmos3-Super-Image2Video — **not affiliated with NVIDIA**.

- **Website**: https://cosmos3ai.net/
- **Model variant**: Cosmos3-Super-Image2Video
- **License**: NVIDIA OpenMDW-1.1 (commercial use with "Built on NVIDIA Cosmos" attribution)
- **Support email**: support@cosmos3ai.net
- **Media CDN**: https://cfsource.cosmos3ai.net/cosmos3super/home/

---

## Homepage (`app/page.tsx`) — what is actually rendered

The public homepage is one continuous landing flow. As of the current `page.tsx`, these sections are **mounted in this order**:

1. **Hero** (`#hero`) — Badge: *Cosmos3-Super-Image2Video · May 31 2026*. **H1**: *Cosmos 3 Super AI Video Generator* (with *AI Video* in green). Lead copy: online generator powered by NVIDIA's 64B open-weights Cosmos3-Super-Image2Video; upload one image, describe motion, get MP4 in browser — no GPU, no install. Background: full-viewport hero video from CDN. **CTAs**: "Animate with Cosmos 3 Super →" (disabled) · "See Example Outputs" (links to `#examples`). **Trust strip**: No GPU required · Commercial use allowed · 1–7 second output · OpenMDW-1.1 license. **Stat cards**: 64B model · Physics-IQ #1 · Max 7s ($0.05/s) · 5 aspect ratios (16:9 · 9:16 · 1:1 · 4:3 · 3:4).
2. **What Is** (`#what-is`) — Label *About the Model*; **H2**: *What Is Cosmos 3 Super Image to Video?* Explains MoT Reasoner + Generator pipeline, OpenMDW-1.1 commercial license, and fact cards (release date, architecture, benchmarks, training data). Sticky **Technical Specifications** panel: model variant, 64B parameters, Physics-IQ #1, 1–7s output, JPG/PNG/WEBP input, OpenMDW-1.1 license.
3. **Video Showcase** (`#examples`) — Label *Output Examples*; **H2**: *Cosmos 3 Super Image to Video — Real Outputs*. Six click-to-play clips with copyable motion prompts (lazy-loaded posters, optional sound toggle). Media from `cfsource.cosmos3ai.net/cosmos3super/home/cosmos-3-super-showcase-{1-6}.mp4`.
4. **Features** (`#features`) — Label *Core Capabilities*; **H2**: *What Makes Cosmos 3 Super Image to Video Different*. Six capability cards (see below).
5. **Use Cases** (`#use-cases`) — Label *Applications*; **H2**: *Who Uses Cosmos 3 Super Image to Video*. Four cards with top video previews (see below).
6. **Comparison** (`#comparison`) — Label *Model Comparison*; **H2**: *Cosmos 3 Super Image to Video vs Alternatives*. Table vs Kling 3 I2V, Wan 2.1 I2V, Runway Gen-4 I2V, Luma Ray 3 I2V (see below).
7. **FAQ** (`#faq`) — **H2**: *Cosmos 3 Super Image to Video — Frequently Asked Questions*. Accordion — see FAQ section below.
8. **Bottom CTA** — **H2**: *Animate Your Images with Cosmos 3 Super*. Primary "Generate Video Now →" button is **disabled** (coming soon).

**Not on the homepage in the current `page.tsx` (commented out):** `Generator`, `HowToUse`, `PricingSection`, `CrossSell`. Do not describe those blocks as live homepage content unless the route is updated.

---

## Core capabilities (`#features`) — card titles on the page

1. **Physics-Real Motion Simulation** — Reasoner analyzes material properties and physics before generation; #1 on open Physics-IQ benchmark.
2. **Image-Guided Video Generation** — Purpose-trained for image input; preserves subject, palette, and composition without hallucinating new elements.
3. **Prompt-Based Motion Control** — Camera moves, subject behavior, atmosphere, lighting transitions; supports negative prompting.
4. **Ambient Sound Generated Automatically** — Omnimodal model renders synchronized ambient audio in the same forward pass as video.
5. **Five Platform-Native Aspect Ratios** — 16:9, 4:3, 1:1, 3:4, 9:16 rendered natively (not cropped).
6. **Commercial Use on Every Plan** — NVIDIA OpenMDW-1.1 license; attribution required: "Built on NVIDIA Cosmos."

---

## Showcase samples (`#examples`) — 6 image-to-video clips

Each tile shows I2V badge, click-to-play video, scene title, motion prompt, and Copy Prompt button:

1. **Mars Humanoid Robots** — I2V · 16:9 · 5s
2. **Desk Robot Chess Move** — I2V · 9:16 · 5s
3. **Neon Alley Delivery Bot** — I2V · 1:1 · 3s
4. **Bimanual Robot Cutting Cake** — I2V · 16:9 · 7s
5. **Airplane Cockpit Landing** — I2V · 9:16 · 5s
6. **Jet Cockpit Maneuvers** — I2V · 16:9 · 5s

---

## Use cases (`#use-cases`) — four cards on the page

1. **01 / E-commerce & Product** — Turn Product Photos into Video Ads
2. **02 / Social Media Content** — Animate Static Assets for Reels and TikTok
3. **03 / Advertising & Agencies** — Produce Multiple Creative Variants Fast
4. **04 / Pre-visualization & Concept** — Animate Storyboard Frames Before Shooting

Each card includes a top video preview from `cosmos-3-super-usecase-{1-4}.mp4` on the CDN.

---

## Comparison table (`#comparison`) — Cosmos 3 Super I2V vs alternatives

| Dimension | Cosmos 3 Super I2V | Kling 3 I2V | Wan 2.1 I2V | Runway Gen-4 I2V | Luma Ray 3 I2V |
|-----------|-------------------|-------------|-------------|------------------|----------------|
| Open Weights | ✓ Yes | ✗ No | ✓ Yes | ✗ No | ✗ No |
| Physics-IQ Rank | #1 open models | — | — | — | — |
| Ambient Audio (same pass) | ✓ Built-in | ✗ | ✗ | ✗ | ✗ |
| Commercial Use (all plans) | ✓ OpenMDW-1.1 | Paid tier only | ✓ Apache 2.0 | Paid tier only | Paid tier only |
| No GPU / Browser-Based | ✓ On this site | ✓ | Self-host only | ✓ | ✓ |
| Negative Prompt Support | ✓ Native parameter | Limited | ✓ | Limited | ✗ |
| Max Duration | 7 seconds | 10 seconds | 5 seconds | 10 seconds | 9 seconds |
| Aspect Ratios | 16:9 · 9:16 · 1:1 · 4:3 · 3:4 | 16:9 · 9:16 · 1:1 | 16:9 · 9:16 | 16:9 · 9:16 · 1:1 | 16:9 · 9:16 · 1:1 |

---

## FAQ (`#faq`) — questions as shown on the homepage

1. What is Cosmos 3 Super Image to Video?
2. Do I need a GPU to use Cosmos 3 Super Image to Video?
3. Can I use the generated videos commercially?
4. What aspect ratios does Cosmos3-Super-Image2Video support?
5. What makes Cosmos 3 Super Image to Video better than other AI video generators?
6. What image formats can I upload to Cosmos 3 Super Image to Video?
7. Will my images or prompts be used to train the model?

**Themes covered in answers (for retrieval):** NVIDIA Cosmos3-Super-Image2Video post-trained I2V model; 64B MoT (32B Reasoner + 32B Generator); #1 open Physics-IQ; cloud/browser-based — no local GPU; OpenMDW-1.1 commercial license with attribution; aspect ratios 16:9, 4:3, 1:1, 3:4, 9:16; JPG/PNG/JPEG/WEBP RGB input at 256p/480p/720p; ambient audio in same pass; uploads not used for retraining.

---

## Technical specifications (from What Is section)

| Spec | Value |
|------|-------|
| Model variant | Cosmos3-Super-Image2Video |
| Parameters | 64B (32B Reasoner + 32B Generator, open weights on HF) |
| Physics-IQ rank | #1 open-weights model |
| Output duration | 1–7 seconds |
| Billing | $0.05 per second of selected duration |
| Input formats | JPG · PNG · WEBP (RGB 8-bit · 256p / 480p / 720p) |
| Max upload size | 10 MB (on this site) |
| License | OpenMDW-1.1 — commercial use with attribution |
| Release | May 31, 2026 — GTC Taipei / Computex 2026 |
| Training data | 1.3B+ data points across text, image, video, and audio |

---

## Pricing (from Hero stat cards)

- **$0.05 per second** of selected output duration (1–7 seconds)
- Example: 5-second clip = $0.25; 7-second clip = $0.35
- A `/pricing` route exists in the codebase but is **not mounted** on the homepage in the current `page.tsx`

---

## Key pages & anchors

- [Home](https://cosmos3ai.net/)
- [Hero](https://cosmos3ai.net/#hero)
- [What Is](https://cosmos3ai.net/#what-is)
- [Output examples](https://cosmos3ai.net/#examples)
- [Features](https://cosmos3ai.net/#features)
- [Use cases](https://cosmos3ai.net/#use-cases)
- [Model comparison](https://cosmos3ai.net/#comparison)
- [FAQ](https://cosmos3ai.net/#faq)
- [Terms of Service](https://cosmos3ai.net/subsidiary/terms-of-service)
- [Privacy Policy](https://cosmos3ai.net/subsidiary/privacy-policy)
- [Refund Policy](https://cosmos3ai.net/subsidiary/refund-policy)
- [Cookie Policy](https://cosmos3ai.net/subsidiary/cookie-policy)

---

## Target users

E-commerce teams, social creators, advertising agencies, and pre-visualization directors who need **physics-real AI video from a single still image** — especially workflows where **open weights**, **native ambient audio**, **browser-based generation (no GPU)**, and **OpenMDW-1.1 commercial licensing** matter.

---

## Compliance & legal

- **Commercial use**: Covered under NVIDIA **OpenMDW-1.1** on all plans; attribution required: "Built on NVIDIA Cosmos." Review this site's **Terms of Service** and **Privacy Policy** for platform-specific rules.
- **Disclaimer**: cosmos3ai.net is an independent resource — **not** the official NVIDIA corporate site.

---

## Contact

- **Support**: support@cosmos3ai.net
- **FAQ**: https://cosmos3ai.net/#faq

---

## Sitemap

https://cosmos3ai.net/sitemap.xml