✕
Display Advanced Options?
⇚
⚙
+ 0/75

Generate
Generate Forever
Generate Previews
Interrupt Current Session
Interrupt All Sessions
Interrupt And Generate
Show Prompt Tokenization
Generate
Generate Forever
Generate Previews
Interrupt And Generate
Interrupt Current Session
Interrupt All Sessions
Quick Tools ⮟
Reset Page Layout
Reload Parameter Values
Reset Params to Default
Open Empty Image Editor
Interrupt Current Session
Interrupt All Sessions
Reshow Welcome Message
Hide Unaltered Parameters
⤋
Model:
Edit Preset
Delete Preset

Seems like the ComfyUI workflow editor direct-access hasn't loaded. Give it a second to load, or try clicking the "Comfy Workflow" tab header - if it doesn't load, there might be an error message in your browser console or the server logs.

Utilities
The "Utilities" tab provides a variety of general utilities and tools in the subtabs above, and the quicktools below.

Metadata Utilities

If you click this button, all Model and Image metadata datastores will be reset, and they will be reloaded from source files.
This is useful for example if you've externally modified your model or image files and want to clean out or update Swarm's metadata tracking.
Note that this does not remove metadata from the models or images themselves, just from the databases that keep an efficient tracker of them.
This may take a moment to run.

Model Sub-type:  Date range:
Filter:  Replace:
Model filename filter:


If you click this button, Swarm will scan Civitai for metadata on models, and update the metadata database with any information found.
You can filter by date range (eg only recently downloaded model files), and/or by required missing content (eg only scan those without images).
You can also choose what metadata gets updated (eg only the image).
Optionally use the model filename filter to only scan models with a specific filename. Use '*' as a wildcard.
(For example, 'SDXL/*' will scan only models within a folder named 'SDXL/'. This is case-insensitive.)
Will take a while, and may seem to freeze up if it has to apply edits. This is normal.
Additional debug info while running can be found in the browser console.
Be careful! If you have manually edited metadata, you can easily accidentally delete it. There is no undo for unwanted changes, they are applied rapidly in bulk.
CLIP Tokenizer
This is a tool to analyze CLIP tokenization for Stable Diffusion models.
All current Stable Diffusion models use the same CLIP token set, so this applies to them all.
Simply type some text in the box below to see how it gets tokenized.
It will show each text-piece, and its numerical ID.



Pickle To Safetensors
This is a tool to quickly convert legacy Pickle (.pt, .ckpt, .bin) files to modern Safetensors files.
WARNING: Pickle files may contain malicious code. Do not use this tool or otherwise load pickle files unless you trust their source.
The pickle files will be moved to a "backups" folder, and the safetensors files left in place where the pickles originally were.
You may delete the backups folder after confirming the new safetensors files work as intended.
Be aware that this may temporarily use a large amount of filespace.
This may take some time to process.
(You can continue using the UI as normal while this runs)
(This will save filespace with minimal side effects, highly recommended)
TypeCountConvert
Models?
LoRAs?
VAEs?
Embeddings?
ControlNets?

LoRA Extractor
This is a tool to extract a LoRA from the difference between two models.
"Base" should be whatever the common base model is, eg SDXL-1.0-Base.
"Other" should be the unique model with information to extract into the LoRA.
The closer Base is to Other, the less complex the LoRA's data will be, and the more likely it will work well with other models that were built off the same base.

Note that LoRA Extraction is an imperfect partial process, and will lose some of the data that makes the model unique.
Rank is a number indicating how much detail to try to save. Higher numbers result in bigger files, and only slightly more accurate matching. Small values (eg 16) are usually sufficient.
Base model:
Other model:
Rank:
Save as:


Model Downloader
This is a tool to download models for you and save them to Swarm's model folder.
Note that only .safetensors or .gguf models may be downloaded.
HuggingFace file URLs are supported.
Civitai URLs are supported and will automatically download+apply model metadata from the civitai API.
For anything else, make sure your URL is a direct download URL (not eg an html page with the info about it).
Exact web URL to the model.
Civitai URLs are fully supported and will load metadata.
HuggingFace URLs will be recognized.
For anything else, make sure the URL is a direct download URL, not a webpage.
Note only '.safetensors' or '.gguf' models are allowed.
? URL
This section will update after you input a URL to inform you if the URL is valid.
The box below will display loaded metadata if you use a civitai URL.
? Status (...)
The base model sub-type to use, ie what category it gets saved to.
Make sure this is correct, you don't want a LoRA in your base models folder or similar.
Civitai links will autoset this when they load metadata.
? Model Type
What folder to put the model in.
To create new folders, simply type paths with '/' in them into the 'Save as:' name box below.
? Folder
File name to save as. Don't include the '.safetensors' file extension, just the name.
You can use a folder path with the '/' symbol.
? Save as
User Account

User accounts are not enabled. You are the local default user. You can enable multi-user accounts in the Server Configuration tab.


API Keys
To use remote APIs in SwarmUI, you must set the key for it. If you have an API key for any of the below APIs, paste it into the corresponding box and click 'Save'.
APIStatusActionsLinkInfo
Hugging Face (Unknown) Go here to create a key If you plan to use the Model Downloader utility to download content from Hugging Face, you may want to set your Hugging Face API key below.
This will allow you to download gated or private content that your HuggingFace account has access to.
Civitai (Unknown) Go here to create a key If you plan to use the Model Downloader utility to download content from Civitai, you will want to set your Civitai API key below.
This will allow you to download gated or early-access content that your Civitai account has access to.

Notice: this is raw internal configuration of parameters. Don't mess with this unless you know what you're doing.

Save 0 edited parameter setting(s)?
Generate Tab Layout Configuration

The settings below let you reorganize the sub-tabs within the Generate tab.
Any sub-tab group that does not have any sub-tabs inside of it will be hidden automatically.



Mobile/Desktop Layout:

Server
The "Server" tab provides access to SwarmUI's server internals, including:
• The Backends tab, which allows you to configure and manage backends (the underlying engines that provide the generation core, or remote instances of swarm this instance is connected to)
• The Server Configuration tab, which allows you to configure and manage Swarm's settings (such as models directory, server host port, etc)
• The Extensions tab, which provides ways to manage your Swarm extensions or install new ones
• The Logs tab, which allows you to view the server logs (eg if you get an error, check Logs -> Debug)

This is not to be confused with the User Settings tab, which allows you to configure your user settings (eg save path/format, default VAE choice, etc)
DotNET 10 Missing
You do not seem to have DotNET 10 installed - this will be required in a future version of SwarmUI.
Please install DotNET SDK 10.0 from https://dotnet.microsoft.com/en-us/download/dotnet/10.0
Server Status
Local Network
This server is only accessible from this computer.
Resource Usage
(Loading...)
Connected Users
(Loading...)
Manage Server
Update
(Loading...)
 

Free Memory
You can free up the VRAM usage, or system memory usage (cache) from backends with the two buttons below.
Shutdown
If you want to shut down the server, click the button below.
Add new backend of type: Show Advanced
Save 0 edited setting(s)?
View: Filter:
Extensions
The "Extensions" tab lets you control your installed Swarm extensions, or install new ones.

Installed Extensions

Name Version Tags Author Description Readme License Actions

Available Extensions

Name Tags Author Description Readme License Actions
FaceTools Parameters, Nodes Quaggles Adds support for CodeFormer FaceRestore and ReActor. Here MIT
WaifuDiffusionV Parameters, Nodes Neggles & Waifu Diffusion Team Adds support for WaifuDiffusion-V-EDM sampling. Here GPLv3
MagicPrompt Tabs, UI Hartsy Provides LLM (language model) usage inside SwarmUI to enhance your prompts. Here MIT
UniversalTab Tabs, UI Hartsy Lets you embed iframes of other webservices into subtabs in SwarmUI. Here MIT
APIBackends api, Backend, Hartsy Integration with commercial image/video generation APIs (OpenAI, BFL, Ideogram, Grok, Google, Fal.ai). Here MIT
PostRender Parameters, Nodes Juan Treminio, orig. HellerCommaA Adds support for a variety of useful post-processing techniques - LUTs, film grain, vignette, blur, and more. Updated and replaced to a new repo March 2026. Here MIT
SkimmedCFG Parameters, Nodes Quaggles Adds support for SkimmedCFG anti-burn allowing much higher CFG scales. Here MIT
SkipLayerGuidance Parameters mcmonkey Adds support for SLG (SD3.5M trick) and STG (LTX-Video trick) as a parameter group. Here MIT
OllamaVision Tabs, UI Urabewe Provides a frontend to connect to Ollama, OpenAI, or OpenRouter for image analysis using vision LLM models. Here MIT
AdaptiveProjectedGuidance Parameters Quaggles Adds support for Adaptive Projected Guidance, retain the quality-boosting advantages of CFG while enabling the use of higher guidance scales without oversaturation. Here MIT
KohyaDeepShrink Parameters derrian-distro A simple SwarmUI extension that adds parameters for the Kohya Deep Shrink / PatchModelAddDownscale node to the generate tab (SDXL resolution trick). Here MIT
TIPO Parameters, Nodes Hugs288 Adds a parameter group for TIPO (anime tag prompt gen). Here MIT
NegPip Parameters, Nodes khanghy1000, orig. Hugs288 Adds a parameter group for Negpip (allows for the use of negative weights in positive prompt, works even in non-CFG models). Here MIT
DAAMSwarm Parameters, Nodes jakstein Provides cross-attention heatmap visualization for Stable Diffusion models. Here MIT
FrameSaver Parameters Mathew Tomberlin Adds parameters to save the first and/or last frame of generated video. Here MIT
VAEReflection Parameters, Nodes Jelosus2 Adds support for using reflect padding mode in VAEs, fixing artifacts in the edges of images generated using a VAE trained with reflect padding, notably EQVAE. Here MIT
MoreThemes UI Blomblo Adds a variety of high-quality themes to SwarmUI Here MIT
PromptBuilder UI, Tools Juan Treminio Adds a new tool UI to browse clickable Danbooru tags to build your prompt. Appears under Tools tab dropdown. Here MIT
SeedVR2Upscaler Parameters, Nodes Rasmus Adds SeedVR2 AI upscaler with auto VRAM detection, quality presets, and 2-step upscaling for reduced artifacts. Here MIT
Base2Edit Parameters, Nodes Juan Treminio Edit an image in the refiner stage (with Edit models like Flux2), allowing you to generate and edit in a single generation. Here MIT
DownloaderHelper Tools Juan Treminio Allows sending model download URLs directly to the SwarmUI Model Downloader page from a "Send to SwarmUI" right-click menu item in Chrome Web Browser. See README for instructions on installing the Chrome extension. Here MIT
VideoStages Tools, Parameters Juan Treminio Allows adding multiple video stages to your workflows, as well as automatically using generate audio. Here MIT
AceStepFun Tools, Parameters Juan Treminio More advanced AceStep 1.5 text-to-audio support for SwarmUI. Works in tandem with the VideoStages to use AceStep audio in LTX2 videos. Here MIT
Quarry Tabs, UI, Tools Juan Treminio Prompt database engine, like wildcards with much more advanced power-user features, supporting Parquet/CSV/JSON/LanceDB datasets, dynamic filters, and more. Here MIT
KreaGoneWild Parameters, Nodes Juan Treminio Adds some neat Krea2 features, like NSFW and Edit-Lora support. Here MIT
SDCppBackend Backend, Beta Hartsy Experimental approach to add a backend powered by an external stable-diffusion.cpp executable. Non-compatible with general use. Here MIT
WildcardImporter Tabs, UI, Low-Quality spoomples+bman654 Lets you import Wildcards from Auto1111-extension-syntax (SD Dynamic Prompts) YAML cards into Swarm's standardizable format. Here MIT
Mahiro Parameters, Low-Quality yoinked Adds a parameter to enable the "MaHiRo" experimental sampling method. Here MIT
WhatTheDuck Tools, Conflicts Juan Treminio Adds potentially-breaking power tools for obscure special cases. Most users should not install this. Here MIT
AltTriggerPhraseBtn Conflicts, UI Blomblo Changes the appearance of the trigger phrase copy button. Here MIT
Epsilon-Scaling Parameters, Nodes, Low-Quality jakstein Adds the Epsilon Scaling node from ComfyUI to SwarmUI as a parameter in the Advanced Sampling group. Here MIT
SharpSplat Tabs, UI, Parameters, Low-Quality Glen Carpenter Turns images into 3D Gaussian Splats directly inside the browser. Here MIT
Spectrum Parameters, Low-Quality Glen Carpenter Adds support for an alternate sampling acceleration technique called 'Spectrum'. Here MIT
WD14Tagger Tools, Parameters, Low-Quality Glen Carpenter Adds automatic image tagging using booru-style taggers. Here MIT
H3Attn Parameters, Nodes, Beta Juan Treminio Attention and speedup options for the MiniMax H3 video model - a built-in window-attention node, plus optional support for the Sol-Attn and Spectrum custom nodes. Here MIT
✕
Roles
Users
SwarmUI v0.9.8.2 (2026-08-25 02:01:07)