Tool not found.
Back to Home
Open-source image and video generation framework for efficient high-resolution AI content creation.
SANA is an open-source generative AI framework developed by NVIDIA Research for efficient high-resolution image and video generation. Built around Linear Diffusion Transformers and advanced compression techniques, SANA delivers high-quality image synthesis while significantly reducing computational requirements compared to traditional diffusion models. The project includes a growing ecosystem of models and research initiatives including SANA, SANA-1.5, SANA-Sprint, SANA-Video, LongSANA, SANA-WM, and Sol-RL. It supports text-to-image generation up to 4K resolution, text-to-video generation, image-to-video workflows, controllable world modeling, reinforcement learning post-training, and fast one-step image generation. SANA integrates with Hugging Face Diffusers, ComfyUI, ControlNet, LoRA, DreamBooth, and SGLang, making it suitable for researchers, developers, AI creators, and organizations building custom generative AI applications. The framework also supports 4-bit and 8-bit quantization, allowing deployment on consumer GPUs with less than 8GB of VRAM. Released under the Apache 2.0 license, SANA provides complete training, inference, evaluation, and deployment pipelines for modern multimodal AI generation workflows.
