Nvidia has unveiled Sana, an innovative AI model designed for instant 4K image generation on ordinary PCs. This model operates using a deep compression autoencoder that reduces image data to a mere 3% of its original size, while maintaining quality. By integrating Google’s Gemma 2 LLM for prompt encoding and decoding, and a Linear Diffusion Transformer to optimize calculations, Sana delivers images four times larger than its competitors while consuming significantly less processing power. The results from preliminary tests have demonstrated impressive speeds, with 4K images being generated in under 10 seconds. As Nvidia prepares to release the model's code as open source, it aims to solidify its place in the AI art sector and expand its user base.

Source 🔗