sd-scripts, diffusers (PEFT LoraConfig), musubi-tuner and ai-toolkit each train adapters for overlapping models, and the diffusers LoRA guide itself points to a separate tutorial for loading Kohya/TheLastBen formats. Which key-naming and rank/alpha conventions are interchangeable, and where do conversions silently degrade results?
Why this is still open: Sources describe each tool's own format; none documents cross-tool compatibility or a conformance test.