HeadsUpAI

fal launches serverless inference for top ranked NVIDIA Cosmos 3 Super

The generative media platform fal has launched serverless inference for NVIDIA Cosmos 3 Super, a 64B parameter omnimodal world model. This integration allows developers to generate high-fidelity video and images by building on the NVIDIA Cosmos 3 release of open weights. The platform supports both image-to-video and text-to-image workflows through a unified API.
Model
Cosmos 3 Super
Parameters
64B
Video Pricing
$0.05 per second
Workflows
Image-to-Video, Text-to-Image
License
Commercial use supported

This launch provides immediate access to the model that currently leads the Cosmos 3 Super analysis for visual generation. It follows a similar pattern to the recent Grok Imagine Video 1.5 launch and joins fal's LTX 2.3 Quality feature in expanding the platform's video generation library.

You can now access cosmos-3-super via the fal playground or integrate it into applications using the serverless API. Video generation is priced at $0.05 per second, with agentic generation billed for each candidate video produced. The model is available for commercial use, supporting high-quality clips with strong motion and multimodal inputs.

View the full update on fal.ai
fal
fal
@fal
X

NVIDIA Cosmos3-Super is now live on fal! Omnimodal generation across image, video and action Image-to-video with strong motion and detail High-quality, temporally coherent clips from rich multimodal inputs https://t.co/zat45hxtRM

8retweets53likes
View on X

Still wondering? A few quick answers below.

NVIDIA Cosmos 3 Super is a 64B parameter omnimodal world model designed for high-fidelity visual generation. It can generate video, images, audio, and action commands from multimodal inputs like text and images. The model currently holds the top rank on open-weights leaderboards for both image and video generation quality.

Video generation using the Cosmos 3 Super model on fal is priced at $0.05 per second of generated content. Costs are rounded up to the nearest second. For agentic workflows where multiple candidate videos are generated, fal bills for each candidate produced during the task execution.

Yes, the NVIDIA Cosmos 3 Super model hosted on fal is available for commercial use. Developers can integrate the model into production applications via the fal API, allowing for the generation of high-quality, temporally coherent video clips and images for commercial products and services.

The fal platform currently supports two primary workflows for the Cosmos 3 Super model: text-to-image and image-to-video. Users can access these capabilities through a web-based playground for testing or via a production-ready API for automated integration into external software and agentic systems.

Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →

Share this update