A live coding session focused on building 'Molina', a new Go library that wraps the stable-diffusion-cpp C++ libraries using FFI (without CGo) to add text-to-image and eventually image-to-video support to the Kron AI platform. The session covers architectural planning, understanding LLM-conditioned diffusion models vs. classic CLIP/T5 encoders, safe tensor file management, model bundle curation strategy, GPU backend support (CUDA, Vulkan, ROCm), and using the Amp AI coding agent to scaffold the initial Go project structure mirroring existing projects Ema and Bucky. Token usage with Claude via Amp reaches over 13 million tokens during the session.
•2h 19m watch time
1 Impression