Running AI image generation locally on CPU only — what actually works in 2025/2026?

Hey everyone, I need to run AI image generation fully locally on CPU only machines. No GPU, minimum 8GB RAM, zero internet after setup. Already tested stable-diffusion.cpp with DreamShaper 8 + LCM LoRA and got ~17 seconds per 256x256 on a Ryzen 3, 8GB RAM. Looking for real world experience from people who actually ran this on CPU only hardware: What tool or runtime gave you the best speed on CPU? What model worked best on low RAM? Is FastSD CPU actually as fast as claimed on non-Intel CPUs like AMD? Any tools I might be missing? Not looking for "just buy a GPU" answers.