Stable Video 3D is a novel multi-view synthesis and 3D generation pipeline that produces detailed and high-fidelity 3D reconstructions from input images.

Scaling Rectified Flow Transformers technique fine-tunes language models with preference data, which results in improved sample quality and spelling.

DistriFusion uses distributed parallel inference to generate highres images, dividing the model input into multiple patches and distributing them across different GPUs.