$ show xgen-universe/Capybara
Capybara is a unified visual creation model that provides inference capabilities for text-to-video, text-to-image, instruction-based video-to-video, and instruction-based image-to-image tasks. The system features a modular architecture with separate components for model handling, pipeline orchestration, and utility functions. Key entry points include the main inference script, ComfyUI custom nodes, and batch/single inference scripts. The system supports distributed multi-GPU inference, prompt rewriting using Qwen3-VL, and FP8 quantization for reduced memory usage.
The LLM identified 2 files worth highlighting for reuse:
Contains reusable utility functions for resolution and bucket calculations, aspect ratio handling, and sophisticated instruction rewriting with separate prompts for editing vs generation tasks. Demonstrates domain-specific expertise in video generation requirements and provides self-contained functions that could be extracted for use in other visual generation projects.
Implements flash attention utilities without padding, handling both flash_attn and flash_attn_interface versions with appropriate fallbacks. Provides a self-contained solution for efficient attention computation that could be reused in other transformer-based projects requiring memory optimization and compatibility across different flash attention implementations.