Pyramid Flow
Open-source AI video generator that runs locally with miniFLUX or SD3 models, supports Apple Silicon, and includes a browser interface.
Pyramid Flow is an open-source AI video generator for people who want to create clips on their own hardware, and for researchers working on video models. It turns text prompts into video or animates a supplied image with guidance from text. The Python code uses PyTorch and carries MIT. Model licenses are separate: the SD3-derived weights use the Stability AI Community License. Check the chosen checkpoint before deployment.
The model can produce clips up to 10 seconds long at 768p and 24 frames per second. Text-guided image animation supports five-second clips. You can choose between miniFLUX and SD3 model variants; miniFLUX focuses on better human anatomy and more stable motion. A local Gradio interface lets you work in a browser, while Hugging Face Spaces and Google Colab provide cloud alternatives.
Hardware needs matter here. CPU offloading reduces GPU memory use to below 12GB, or below 8GB with sequential offloading, at the cost of speed. Apple Silicon can run the model through the MPS backend. Multi-GPU inference for the SD3 variant spreads the workload across two or four GPUs to reduce memory use per GPU and speed up generation.
Its distinguishing approach is pyramidal flow matching: generation works across different resolutions and noise levels rather than doing all the work at full resolution. The model generates video progressively and was trained using only open-source datasets. The repository also includes training and fine-tuning code, with at least eight A100 GPUs required for VAE training or DiT fine-tuning.
