SwarmUI
SwarmUI is a modular Stable Diffusion web interface focused on performance, extensions, batch generation and multi-GPU workflows through a local browser.
What SwarmUI is for
SwarmUI brings models, prompts, sampling parameters, batch jobs and extensions into a browser interface backed by a local service. It is aimed at users who already understand checkpoints, memory limits and generation backends, especially when grid generation or multi-GPU work is useful. A cloud service may be simpler for occasional image generation.
Installation and hardware boundaries
Windows, Linux and Apple Silicon macOS use different scripts and runtime combinations. Git, .NET, Python, a generation backend and a compatible driver may all be involved, and first launch can continue downloading components or models. Intel Mac support and nonstandard GPU backends should be treated as limited until tested.
Network and extension boundaries
The browser view does not make the service a public website. Keep the listener on the local machine unless an authenticated, restricted reverse proxy is deliberately configured. Models, workflows and extensions can carry separate licenses or execute extra code, so the main project license does not describe every installed component.
Maintenance note
This page was reviewed on 2026-08-23 against the 0.9.8-Beta setup notes, runtime requirements and MIT license. Beta dependencies can move quickly, so keep the extracted package and model list together when troubleshooting.
Save to your cloud drive
Open the cloud drive to get the file directly, or save it for convenient access on another device.
Quark Cloud Drive
RecommendedSave SwarmUI to this cloud drive
Baidu Netdisk
Save SwarmUI to this cloud drive
SwarmUI installation and first local generation
Install SwarmUI 0.9.8-Beta, configure a compatible backend and generate one small test image before importing extensions or large models.
Before you start
- Prepare Git and the runtime versions required by the selected operating system and backend.
- Check graphics driver support, available VRAM, disk space and network access for model downloads.
- Keep the service directory writable by a normal user and reserve a separate output directory.
Installation steps
- 01
Choose a local installation directory
Use an independent directory writable by the current user. Avoid protected system folders and avoid running the service as an administrator for everyday use.
- 02
Run the system-specific launcher
Start the Windows, Linux or macOS script from a terminal and retain the output. Complete logs are more useful than repeatedly double-clicking a failed script.
- 03
Finish the local setup page
Open the local browser page, select a backend and model directory, and wait for dependency processing to finish before closing the terminal.
Quick start
- 01
Load one compatible model
Choose a fully downloaded checkpoint that matches the selected backend, and confirm that its configuration and weight files are complete.
- 02
Generate a small test image
Use a simple prompt, one image, a modest resolution and default sampling settings to verify the pipeline, VRAM and output directory.
- 03
Save reproducible parameters
Record the model, prompt, negative prompt, seed, size and sampler before changing one setting at a time.
Usage tips
- Keep the local service bound to the local machine; remote use needs authentication, a restricted network and an appropriate proxy.
- Review the source and license of every model, extension and workflow before importing it.
- Reduce resolution, batch size and parallel jobs before switching to a lower-precision backend when VRAM is tight.
Troubleshooting and uninstall
Why did the browser page not open automatically?
Read the terminal for the local address and port, then open that address manually. If the port is occupied, stop the old instance or choose another port instead of starting multiple copies.
Why does the launcher close with a missing runtime message?
Run it from a terminal and inspect the complete error, then check Git, .NET, Python versions and directory permissions before trying again.
- Back up models and workflowsCopy models, generation parameters, workflows and outputs that should be kept, and remove private prompts from files that will be shared.
- Stop the service and remove the directoryStop the process from the terminal, then remove the installation directory after confirming that its models are not shared with another interface.
Frequently asked questions
Is SwarmUI a standalone desktop window?
It starts a local service and is operated in a browser. Closing the browser does not necessarily stop the service; end the process from its terminal or process manager.
Can SwarmUI run without a discrete GPU?
Some integrated or alternative backends may run with reduced speed or features. Test a small workflow and confirm backend and driver support before downloading large models.
Should the local port be exposed publicly?
Keep it local by default. Any remote setup should add authentication, a restricted network boundary and a carefully configured reverse proxy.
Does MIT cover every model and extension?
No. Check each model, workflow and extension license and review whether it executes extra code or sends data outside the local machine.