After the overwhelming feedback I received for my second project, MyLoRA, I figured it’s time to let you all in on the next experiment: SDUnity.
Let’s face it – many current T2I/I2I frameworks are stuck in the past. A1111, Fooocus, ComfyUI, even Forge... all still run on outdated components like Gradio 3, Python 3.10, or legacy Torch versions. Some are slowing down development or have even gone dormant. So I decided: Let’s build something from scratch. Modern. Clean. Focused.
The result?
SDUnity – a work-in-progress, but already functional self-hosted interface for Stable Diffusion that puts everything in one place: generation, model management, tagging, gallery, Civitai browsing, and more. Built with Gradio 5 and up-to-date libraries across the board.
💡 Looking for collaborators! If you’re interested in improving SDUnity, building features or refining the UI/UX – feel free to reach out. Contributors welcome!
🧠 Why SDUnity?
The goal of SDUnity is to streamline the chaotic AI art ecosystem. It’s a single tool where you can:
Prompt & generate images,
Manage models and LoRAs,
Auto-tag content,
Browse and import from Civitai,
Organize everything with a built-in gallery.
It’s fast, focused, and designed with modern workflows in mind. Just enough features to be powerful, without getting bloated.
🚀 Features at a Glance
🎨 Generator
Full prompt and negative prompt control
Seed management (fixed/random)
Sampler selection, CFG, steps, width/height
Clip skip toggle, precision mode, tiling
Parallel generation & batch repetition
Smooth step streaming preview (WIP)
High-res fix & denoise strength
LoRA weight sliders
Wildcard support via
__name__
tokens (wildcards/
)GPT‑2 based Prompt Enhancer (adds quality/detail/preset tags) (WIP)
NSFW filter toggle
📦 Model Manager & Civitai Browser (🔥 API-Powered)
Managing your models has never felt this smooth. The Model Manager in SDUnity lets you:
Browse all local models under
models/
and LoRAs underloras/
Move, categorize, or delete files
Load/unload pipelines from VRAM
But the real game-changer?
Full Civitai API integration.
You can:
Search directly from within the UI (SD 1.5, SDXL, Pony tags included)
View model versions, metadata, and previews
Download models and LoRAs with a single click
Or fetch them via terminal using the CLI tool
All without ever leaving SDUnity
This feature is one of the smoothest parts of the system and something I'm personally very proud of. It just works.
🖼️ Gallery (WIP)
Browse saved generations from
generations/
Full metadata viewer (prompt, seed, LoRA, model, etc.)
Delete unwanted images directly from the UI
✨ Prompt Tag Suggestions
Inline autocomplete using fuzzy search (
rapidfuzz
)Type in the prompt field and press Tab to complete
Powered by a local dataset (rebuilt via maintainer script)
Based on
a1111-sd-webui-tagcomplete
, rebuilt withscripts/import_tagcomplete.py
🏷️ Image Tagger
Tag anime-style images automatically using WD14
Fully local – no web request or upload needed
Tags saved into metadata for reuse, filtering or searching
🧰 Setup (Linux – Windows may work)
git clone https://github.com/AsaTyr2018/SDUnity.git
cd SDUnity
sudo ./maintainer.sh install
Start the interface:
/opt/SDUnity/start.sh
Default web UI: http://localhost:7860
All configs can be adjusted via the Settings tab or in sdunity/config.py
.
⚠️ Windows Support: The core components are cross-platform, and SDUnity should work on Windows (e.g. via WSL or native Python installs). However, it has not been fully tested yet. Community testers are welcome!
🛠 Maintainer Script
Use maintainer.sh
for everything:
sudo ./maintainer.sh install # First setup
sudo ./maintainer.sh update # Pull new changes
sudo ./maintainer.sh uninstall # Remove everything
It handles the venv, dependencies, and updates cleanly. You only need git
and python3
.
📌 Notes
Prompt presets live in
presets.txt
⚠️ Status: Active Development
Let’s be real: SDUnity is still in early development. Some features are WIP, some polish is missing – but the core is already rock solid and fun to use. If you’re tired of legacy UIs or want something fresh that keeps pace with the AI ecosystem... SDUnity might be what you’ve been looking for.
Check out the repo here:
👉 github.com/AsaTyr2018/SDUnity
Feedback, testers, feature ideas – and especially contributors – are always welcome.
This is just the beginning.
Previews:





