Capybara: A Unified Visual Creation Model
Summary
Capybara introduces a unified visual creation model built on diffusion and transformer architectures, enabling multi-task generation and editing (T2I, T2V, TI2I, TV2V) with high-performance distributed inference. The project highlights FP8 quantization, ComfyUI integration, and both single-sample and batch inference modes, with updates in early 2026.