Mim is a privacy-first experimental SNS that shares the “meaning” of images instead of pixels. Images are never uploaded; on-device AI extracts semantic data, and the receiver reconstructs visuals locally using Stable Diffusion. This enables low-data, private visual communication. Reconstructed images may differ from originals. Features include privacy modes, Pro mode (fidelity, time, memory), social functions, and full on-device processing. Open source, built with Swift and Core ML.