Skip to content

Overview

Effective prompting is foundational to getting the best results from Cosmos models. This section provides comprehensive guides on how to interact with Cosmos models through well-structured prompts.

Available Guides

Cosmos Reason 2 Prompt Guide

A comprehensive guide to prompting Cosmos Reason 2, covering:

  • Message structure and media ordering
  • Sampling parameters and their effects
  • Domain-specific prompting patterns
  • Best practices for different task types

This guide helps developers, researchers, and practitioners understand how system prompts, user instructions, sampling parameters, and multimodal message structure influence model behavior across vision, robotics, autonomous driving, and physical-world understanding tasks.

Key Concepts

  • Prompt Structure: How to organize system prompts, user instructions, and media inputs
  • Sampling Parameters: Tuning temperature, top-p, and other parameters for different tasks
  • Structured Outputs: Requesting specific output formats like JSON, trajectories, or reasoning chains
  • Media Ordering: Proper sequencing of images, videos, and text in multimodal inputs

These guides consolidate best practices and common patterns to help you design effective workflows and integrate Cosmos models into real-world systems.