// the find
MochiDiffusion/MochiDiffusion
Run Stable Diffusion on Mac natively
Native macOS app for running Stable Diffusion and FLUX.2 Klein locally using Apple's Core ML framework and Neural Engine. Targets Apple Silicon Macs specifically — this is not a cross-platform port. For Mac users who want offline image generation without Python environment hell.
1. Actually uses the Neural Engine properly via Core ML split_einsum models, getting ~150MB memory usage that a PyTorch approach would never achieve. 2. EXIF metadata baked into generated images with prompt info — that's a detail most tools miss and it's genuinely useful for managing outputs. 3. Clean separation of model types (SD via Core ML, FLUX.2 Klein via iris.c submodule) rather than trying to shoehorn both through one pipeline. 4. Localized into 15+ languages with Crowdin integration — unusual commitment for a niche tool.
1. Requires macOS 15.6+ and Xcode 26.0 to build — these are unreleased as of mid-2026, so the compatibility table is aspirational, not current. Anyone building today will hit version mismatches. 2. Model setup is manual and fiddly: download, convert via Python script, place in exact directory structure, then wait 2 minutes for Neural Engine compilation on first run. No in-app model browser or one-click downloads. 3. FLUX.2 Klein support leans on iris.c as a git submodule with a bridging header — that's a C library bolted on via Objective-C interop, which will be a maintenance headache as both projects evolve independently. 4. No batch prompt processing or scripting interface — everything is interactive GUI only, so automated workflows are a dead end.