GauGAN2
by NVIDIA
Create photorealistic landscapes from sketches and words using NVIDIA's GAN
Visit Product
131 upvotes
644 views
About
GauGAN2 is NVIDIA's AI art tool that creates photorealistic landscape images from rough sketches, text descriptions, or a combination of both. Building on the original GauGAN (Generative Adversarial Network for creating photorealistic scenes), GauGAN2 adds multi-modal input support — allowing users to describe a scene in words and then refine it with brush strokes, or draw a layout and have text descriptions influence the style and content.
The system's ability to seamlessly blend text and sketch inputs makes it unusually versatile: a user might start with 'snowy mountain at sunset' and then sketch in a specific tree placement, or draw rough shapes and use text to specify what each region should contain. The AI synthesizes all these inputs into a coherent, photorealistic landscape.
GauGAN2 represents a significant research advance in conditional image synthesis and human-AI collaborative art creation. It was developed by NVIDIA Research and demonstrates how AI can serve as a powerful creative partner for artists who want to maintain control over composition while leveraging AI's ability to add photorealistic detail and coherence.
The system's ability to seamlessly blend text and sketch inputs makes it unusually versatile: a user might start with 'snowy mountain at sunset' and then sketch in a specific tree placement, or draw rough shapes and use text to specify what each region should contain. The AI synthesizes all these inputs into a coherent, photorealistic landscape.
GauGAN2 represents a significant research advance in conditional image synthesis and human-AI collaborative art creation. It was developed by NVIDIA Research and demonstrates how AI can serve as a powerful creative partner for artists who want to maintain control over composition while leveraging AI's ability to add photorealistic detail and coherence.
Product Features
- Text-to-image landscape generation
- Sketch-to-photorealistic transformation
- Combined text and sketch input mode
- Segmentation map painting for precise control
- Photorealistic sky, vegetation, water, and terrain
- Real-time generation preview
- Multiple style and weather options
- Interactive refinement with brush tools
- Research demo accessible online
- Resolution up to 512x512 in the demo
- Sketch-to-photorealistic transformation
- Combined text and sketch input mode
- Segmentation map painting for precise control
- Photorealistic sky, vegetation, water, and terrain
- Real-time generation preview
- Multiple style and weather options
- Interactive refinement with brush tools
- Research demo accessible online
- Resolution up to 512x512 in the demo
About the Publisher
NVIDIA Research is the research division of NVIDIA Corporation, the world's leading GPU manufacturer. NVIDIA has been one of the most prolific contributors to AI research, developing foundational work in GANs, neural rendering, and generative models. GauGAN (published as SPADE in 2019) and its successor GauGAN2 represent NVIDIA's contribution to human-AI collaborative image creation — tools that demonstrate both the capability of deep learning and the importance of user control in generative systems.