Sana

Sana is a local AI image and video generation framework for creators and developers who want to generate media on their own hardware. It includes training and inference tools for working with its diffusion models.

• Image generation: Create images from text prompts at resolutions up to 4K, or use SANA-Sprint for generation in one or a few steps.
• Custom models and image control: LoRA and DreamBooth support let you train customized image models, while ControlNet gives you control over image generation beyond the text prompt.
• Video and world generation: SANA-Video generates video, and SANA-WM produces video worlds with control over camera position and rotation.
• Streaming video editing: SANA-Streaming edits incoming video in real time and maintains consistency between frames across minute-scale videos.
• Workflow integrations: Use Sana with ComfyUI for visual workflows or Hugging Face Diffusers for image generation and LoRA training.

The Sana-0.6B image model can run on a laptop GPU with 16GB of memory. With 4-bit quantization, image generation can use less than 8GB of VRAM; those figures don't establish the hardware requirements for its video models. Hosted demos and the Replicate API provide separate cloud options for trying image generation.

The framework is open source under Apache 2.0. Its image models use Gemma to encode text prompts.