Tendem
Mini toolsSana
By NVIDIA
Sana is a highly efficient text-to-image system that combines a 32× deep compression autoencoder, linear-attention DiT and a small decoder-only LLM text encoder. It produces 4096×4096 images with strong text alignment, competitive FID and GenEval scores, and can run a 0.6B model on a 16 GB laptop GPU at under one second for 1024×1024. Flow-DPM-Solver and smart caption selection cut sampling steps and training cost.
New
Image
Gen 4
Overview
Sana is NVIDIA and MIT’s text-to-image framework that uses a linear Diffusion Transformer and a deep compression autoencoder to generate up to 4K images efficiently, matching much larger models while running on a laptop GPU.
About NVIDIA
Industry:
Computer Hardware Manufacturing
Company Size:
36000
Location:
Santa Clara, California, US
Website:
nvidia.com
Last updated: January 14, 2026
