DiffusionNexus

DiffusionNexus is a cross‑platform desktop application for organising Stable Diffusion LoRA models and managing training datasets. It scans your collection, fetches metadata from Civitai and presents each model as a card with preview images. The project uses Avalonia so the same binaries run on Windows, Linux and macOS.

Features

LoRA Model Management

• Thumbnail generation – automatically creates WebP previews from GIF or video files when no static image is present.
• Search & filtering – instant search with autocomplete, folder tree filtering and sort options.
• Metadata download – retrieve missing information from the Civitai API using your API key.
• Duplicate detection – scan any folder for .safetensors files with identical content.
• Clipboard helpers – copy trained words or model names with a single click.

Dataset Management

• Version control – organize training data into versioned datasets with branching support.
• Image captioning – edit and manage captions for training images.
• Batch crop/scale – resize images to standard aspect ratio buckets for LoRA training.
• Image editor – built-in editor with crop, rotate, color balance, brightness/contrast adjustments.
• AI background removal – remove backgrounds using RMBG-1.4 model (runs locally).
• AI upscaling – upscale images using 4x-UltraSharp model (runs locally).
• Rating system – mark images as production-ready or rejected for quality control.

Installation

1. Install the .NET 10 Runtime if required.
2. Download a release archive from the GitHub Releases page.
3. Extract the archive and run DiffusionNexus.UI (on Windows) or dotnet DiffusionNexus.UI.dll on Linux/macOS.

Usage

A full walkthrough is available in the User Guide. Configure paths and API keys under the Settings tab then open Lora Helper to browse your models.

Development

Prerequisites

• .NET 10 SDK
• Visual Studio 2022 or later / VS Code / Rider

Building

dotnet build

Running Tests

dotnet test

Key Dependencies

• Avalonia – Cross-platform UI framework
• FFMpegCore – Downloads platform specific FFmpeg binaries on first run to capture frames from videos.
• SkiaSharp – Image processing, decodes GIFs and encodes WebP images.
• Entity Framework Core – SQLite database for metadata caching.
• ONNX Runtime – Local AI model inference for background removal and upscaling.

Project Structure

Licence

DiffusionNexus is released under the MIT License.

Third-party components are listed with their licences in THIRD-PARTY-NOTICES.txt, which also ships next to the application and is viewable in-app under About. The file is generated from the restore graph by Scripts/Generate-ThirdPartyNotices.ps1 and verified on every pull request — do not edit it by hand.