KrissTos/off-line-Image-gen-mac
Fully offline AI image generation for Mac Silicon (MPS) & NVIDIA. FLUX.2, Z-Image Turbo, LTX-Video — no cloud, no subscriptions.
What's novel
Fully offline AI image generation for Mac Silicon (MPS) & NVIDIA. FLUX.2, Z-Image Turbo, LTX-Video — no cloud, no subscriptions.
Code Analysis
11 files read · 4 roundsA fully offline AI image generation and editing application for Mac Silicon that wraps multiple diffusion models (FLUX.2-klein, Z-Image Turbo, LTX-Video) with a FastAPI backend, React frontend, custom LoRA stacking, quantized model support, depth map generation, watermark removal, and ComfyUI workfl
Strengths
Genuinely useful local-first tool with thoughtful MPS-specific optimizations, a custom LoRA implementation for quantized models that works around PEFT limitations, and a well-designed async pipeline that bridges sync ML code to a streaming React UI. The model management (version comparison, HF cache sync, online version checking) is production-quality.
Weaknesses
app.py is an 86kb monolith that mixes model management, generation logic, and UI state — the core generation loop is buried in a file that's hard to navigate. Test coverage is thin (only 4 test files, mostly for peripheral features like erase and downloads), with no tests for the critical generation pipeline, LoRA loading, or quantized model paths.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
66
Contributors
1
Files
67
Active weeks
2
Repository
Language
Python
Stars
1
Forks
0
License
—