akhaliq/Cosmos3-Nano

Cosmos3 Nano — World Foundation Model

NVIDIA Cosmos 3 Nano: Generate physically plausible images and videos from text, images, and sound using a unified world foundation model.

Features

• —Text → Image — Single-frame generation from text prompts
• —Text → Video — Multi-frame video generation with quality-control negative prompts
• —Image → Video — Animate a conditioning image into video
• —Video + Sound — Generate video with synchronized audio

Architecture

Built with gradio.Server for a custom cinematic frontend with Gradio's queuing backend:

• —Full-screen media display with glassmorphism controls
• —@app.api() endpoints with GPU queuing and concurrency management
• —@spaces.GPU for ZeroGPU allocation on HF Spaces

Model

Uses nvidia/Cosmos3-Nano via the Cosmos3OmniPipeline from 🤗 Diffusers.