Releases: Enuclea/Diffusion4Mac
Releases · Enuclea/Diffusion4Mac
Release list
v2.7.2-beta - Ideogram Local Import Fix
Release resolving quantized local Ideogram 4.0 pipeline loading issue in compiled production builds.
v2.7.1-beta - macOS FLUX & Low-RAM Optimizations
Release containing optimizations for local FLUX models, dynamic FP8 quantization loading, VRAM allocator cache flushes, and image-to-image target dimension overrides.
v2.7.0-beta
Native macOS Apple Silicon support build.
- Native arm64 compilation of the Electron app shell (eliminates Rosetta translation).
- Compiled all native dependencies for arm64.
- Includes the native macOS ARM64 backend.
v2.6.3-beta
Diffusion4Mac v2.6.3-beta Release
- Self-Contained Offline Execution: Bundled with a frozen Python backend, offline Ollama, and RealESRGAN binaries to remove all external system dependencies.
- Apple Silicon MPS Fixes: Resolved the black image bug on macOS by setting the default precision to float16 and upcasting VAE operations to float32.
- Flux Klein 9B CUDA Crash Fix: Patched Qwen3Next in transformers to bypass the CUDA-specific
FusedRMSNormGatedlayer when CUDA is not available on the platform, resolving the startup crash. - Packaging Improvements: Integrated dependency overrides to bypass macOS python2 removal issues in electron-builder.
Beta Release 2.6.2
This beta release fixes LoRA architecture compatibility issues (FLUX.1 Schnell vs FLUX.2 Klein), asset download progress tracking, and adds target dimension capping for img2img to prevent system RAM runaway.
Release v2.6.1-beta
Visual and Performance Optimizations for Diffusion4Mac
This beta release introduces Gemini Studio (cloud generation), performance enhancements, and bug fixes for local pipelines.
Features & Additions
- Gemini Studio: Added a dedicated tab enabling cloud-based image generation via Google Imagen (predict API) and prompt enhancement via Gemini 2.5 Flash. Requires a Gemini API Key.
- Dynamic Model Selection: Automatically lists and discovers available Imagen models dynamically via the Google AI Studio ListModels endpoint.
- Settings Redesign: Arranged settings into a glassmorphic card-grid layout with vibrant color headers.
Bug Fixes & Optimizations
- Img2Img Aspect Ratio Fix: Resolved vertical aspect ratio compression/squashing in
FluxImg2ImgPipelineby explicitly calculating and passing height/width dimensions derived from the preprocessed input image. - Img2Img Strength Setting: Correctly parsed and mapped the user-defined image strength slider value to the Python backend.
- IPC Path Routing Fix: Prevented canvas load failures by passing raw filesystem absolute paths instead of
file://prefixed URLs. - Memory Leak Fix: Added
multiprocessing.freeze_support()to prevent runaway child process recursion in the packaged bundle. - Splash Screen Asset Fix: Fixed HTML images blockages under Electron's custom protocols.
- Read-Only FS Crash Fix: Stopped local debug logging file writes in the packaged app.
BETA v2.3.0
Beta release. YMMV