When GGUF image model fails to load with llama.cpp, try loading with stable-diffusion-cpp-python (sd.cpp) as fallback.