Feature Announcements Thread

[mcmonkey4eva]

This is a dedicated thread you can follow where new features will be announced.

While the project is in alpha, things will likely move fast and there will be posts here somewhat often. After alpha is done, I'll probably open a Releases thread for a slower roundup.

Migration Notice

As of 2024/06/21 StableSwarmUI will no longer be maintained under Stability AI.

The original developer will be maintaining an independent version of this project as mcmonkeyprojects/SwarmUI

Windows users can migrate to the new independent repo by simply updating and then running migrate-windows.bat

For Linux, Mac, or manual Windows: open a terminal in your Swarm folder then run git remote set-url origin https://github.com/mcmonkeyprojects/SwarmUI

See full migration guide here: mcmonkeyprojects/SwarmUI#2

[mcmonkey4eva]

Key feature update: refiner native integration! If you use ComfyUI backend, the refiner stage is now readily supported.

There's a checkbox to download it while you install, and:

You can choose whether to refine, if so with which model, and which method (StepSwap or PostApply), how strongly to refine, and optionally whether to upscale the image (and if so, how).

You can also use mid-run upscaling on its own without swapping the model (aka highres fix)

(This is for the easy-mode without automatic workflow generation, if you use custom workflows this doesn't change, you could already do refiners that way)

[mcmonkey4eva]

When you use a custom Comfy workflow, it now is able to automatically recognize the list of valid inputs - so for example if you have a LoRA loader node, it will show your list of lora files

this also applies to the Grid Generator tool, wherein the Fill button now works to fill the appropriate values

[mcmonkey4eva]

A lot of work has been done on custom comfy workflows lately.

Most notably, you can now title a Primitive input with the format SwarmUI: Prompt to tell swarm to automatically recognize that input as using the default Prompt parameter. You can also do this for any other parameters, as long as the name is correct. (and caps/spacing/etc is ignored, so eg swarmui:seed, Swarmui: comfyui_sampler, etc. all will work)

I use prompt in this exactly as that's commonly broken by using the advanced encoder, but naming the primitive to get it back is very useful for having easy compatibility with presets.

[mcmonkey4eva]

Image History, Presets, and Models (the 3 Browsable parts) now have a 'Depth' input so you can view x many folders of depth, and a Filter text box to only show results that match the filter text (eg you can search "SDXL" in the models area to only show SDXL models) (filter checks name/description/type/etc for models, name/metadata for images, etc - case insensitive)

Also it has lazy-load on images, so if you view deep image history it won't die trying to load the images, it loads as you scroll.

Also there's a cap of 1000 results while deep-viewing images to avoid issues, this will be configurable in the future.

[mcmonkey4eva]

There's now a working interface for LoRAs and VAEs (for generated workflows, doesn't do anything for custom workflows).

LoRAs selected show up in the status bar, with a strength adjustment input.

Just as with the models page, any lora with proper metadata on it will render its metadata

And you can edit the metadata freely via the lil menu button.

(Also a browser for Embeddings that at the moment doesn't do anything functional.)

[mcmonkey4eva]

Custom Comfy Workflows can be saved, loaded, and compared in grids now! (You can also delete them from the Load menu)

[mcmonkey4eva]

There's now a (basic initial) ControlNet option on the main generate tab

[mcmonkey4eva]

You can now quickshut the leftside parameter view with a button press and get a simplified/focused prompt-only view:

Just you, your image, and your prompt. Hit enter to quicksend the prompt. But all the fancy params are still accessible by popping open the left bar at any time.

[mcmonkey4eva]

Initial ReVision native support:

Just slap some images to the prompt box and it'll figure it out for you. Only compatible with Comfy backend + default workflow generator. Will autodownload clip_vision_g for you if you lack it and you're using comfy-self-start and/or at least have your model dir pointed correctly.

[mcmonkey4eva]

There's a css/format rework in-progress. You can git checkout cec01a12c8533d1fc7934cfa9abd99fe0f52a22f to manually go back to before this if it causes any issues for you.

Notably parameters now are less space-heavy, and Prompt/Generate are now in the main center area, so you can just close the parameter list and focus on prompting when you're not in the mood for fiddling settings.

This is still a work-in-progress, there's more changes planned before this is done.

[mcmonkey4eva]

You can now right-click the generate button to enable "Generate Forever", which will queue up a new generate each time the last one finishes, so you can see the results of your changes as you adjust things. Works best with fast models / low step counts / static seed.

You can also right-click the interrupt x button to get "Interrupt All Sessions", so if you accidentally close a window that has a long queue and want it to just stop, you can use that button in a new window to stop it.

Interrupt is also available under Quick Tools.

[mcmonkey4eva]

There's now a tab under User where you can muck with some of the raw parameter configuration (eg number view type / min/max, or example values as are filled by gridgen, or etc)

[mcmonkey4eva]

The Grid Generator tool now has search + tab complete as you type.

You can click a suggestion with your mouse, or use Arrow key up/down to navigate and then hit Tab to use it.

[mcmonkey4eva]

I published a convenient download for the SDXL official preset pack in a Swarm-ready format, including icons and all!
Find that doc here: https://github.com/Stability-AI/StableSwarmUI/blob/master/docs/Presets.md

[mcmonkey4eva]

In my time circling SD-related support channels, I've answered user questions many times about "how does CLIP tokenize this text?" / "does CLIP recognize this symbol?" / "Is this even a word CLIP knows?" / etc. So... Swarm has a utility to answer those questions easily now:

It's under the new "Utilities" tab, and is a relatively basic interface, but works well.

Also! When generating images, any prompt input now comes with a CLIP Token counter that shows how many tokens your current prompt has, and tells you when you hit the CLIP 75-token break line (just like what Auto WebUI has)

[mcmonkey4eva]

New: LoRA Extractor Utility

As it says on the tin: lets you extract LoRAs from your models

[mcmonkey4eva]

New: Tab completion in prompts! Just type a < and you will be suggested all available prompt syntax. Most prompt syntax options have further tab completions once selected.

[mcmonkey4eva]

New! Automatic welcome messages.

Whenever you first open swarm, it will have a welcome message. This can be one of three categories:

• If you're new to swarm (haven't selected a model), it will tell you to select a model. Once you do that, it will switch to telling you to type a prompt below and hit generate. I hope to expand this with more helpful getting-started tips.
• Recent new features selected at random, (to improve awareness of feature updates)
• Random "did you know" tips for interesting or lesser-known features of swarm worth giving a try

These messages occupy the location where generated images would be, and as such go away as soon as you generate an image or pull one up from history.

You can refresh the page at any time to see a new random welcome message.

There is a server setting available (UI -> OverrideWelcomeMessage) to manually set a custom welcome message body if you don't like seeing the generated ones, or if you're sharing a swarm instance and want to show a message to your users.

If you're curious, you can see the full list of defined welcome messages @ https://github.com/Stability-AI/StableSwarmUI/blob/master/src/wwwroot/js/genpage/welcomemessages.js

[mcmonkey4eva]

Grid generator upgrade!

Tools -> Grid Generator got some fun new tricks recently.

• Has a filled out readme now https://github.com/Stability-AI/StableSwarmUI/blob/master/src/BuiltinExtensions/GridGenerator/README.md

• There's now a dropdown to select between Web Page, Just Images, or Grid Image
  
  Naturally, web page is the classic full-detail magic webpage it always had, Just Images will bulk-generate images without making a grid out of them, Grid Image will do a simple single 2D giant grid image at the end like that old auto webui script did

• There's now Save/Load config buttons
  
  These save/load your entire grid setup - output type, folder name (for web page gen), checkbox/settings, and axes/values. This allows you to easily recontinue prior grids on the fly.

• You can now use SKIP: (all caps) before a value to skip it. Useful for the above mentioned reusing-an-old-grid concept.

• There's now a lil hint spot mentioning the new tricks with a link to the new readme about the grid generator. Hover your mouse over the hint to have it light up, otherwise it's faded to not be annoying.
  

• There's now live error checking as you type, automatically detecting invalid inputs where possible to highlight them red. Hover your mouse over a red bubble to see why it was rejected.
  

[mcmonkey4eva]

Installer improvements!

• There's now a clean install script for Linux users.
• The installer UI is not a lot nicer
  • no longer shoves a whole license in your face at the start
  • has a bunch to skip all the configuration steps and just use the defaults, since the defaults are best for most users anyway
  • now shows a progress bar as its downloading files at the end.

[mcmonkey4eva]

New! Translations! The UI is now translated into several common languages, and many more will be added in the coming few days.

I bulk generated the initial translations using an AI model from google (madlad400-7b-mt-bt) using a tool I built to quick bulk that task: https://github.com/mcmonkeyprojects/translate-tool

I welcome PRs to improve the quality of the translations, see also #80 (comment)

[mcmonkey4eva]

Stable Cascade support!

See model support doc here for specifics: https://github.com/Stability-AI/StableSwarmUI/blob/master/docs/Model%20Support.md#stable-cascade

Simply use the "ComfyUI" / "All In One" format models and save them together in your checkpoints folder https://huggingface.co/stabilityai/stable-cascade/tree/main/comfyui_checkpoints

[mcmonkey4eva]

Comfy workflow browser!

Are you a ComfyUI fanatic with hundreds of workflows? Organize them with names/thumbnails/etc. with a quick and friendly UI, directly in Swarm's Comfy tab!
Swap through bowls of spaghetti like someone who's about to get kicked out of the italian all-you-can buffet!

[mcmonkey4eva]

New themes: Modern Dark and Modern Light are now available, with Modern Dark as the new default theme!

Built by Stability AI designer Eric and pythongosssss

... also, for fun, here's the first design idea Eric made, before we started simplifying things down a bit:

(What do you think? Should we have gone with the this "retro control panel" aesthetic? Is there something that could be made better? Feel free to poke me in https://discord.gg/stablediffusion #stableswarmui channel with opinions, or start a discussion in the Discussions tab here on github)

[mcmonkey4eva]

Prompt editing! One of the most requested features - a hit from Auto WebUI, long missing in other UIs, now available in Swarm with comfy backend!
These features are a fun way to mix distinct concepts in one generation.

For alternating words, use <alternate:cat, dog>
For editing prompts between timesteps, use <fromto[0.5]:cat, dog>
See docs here: https://github.com/Stability-AI/StableSwarmUI/blob/master/docs/Prompt%20Syntax.md
also if you have auto webui syntax for this copypasted, like [cat|dog] or [cat:dog:0.5], this will be recognized and work automatically.

[mcmonkey4eva]

Swarm has a dedicated full discord of its own now! https://discord.gg/q2y38cqjNw

[mcmonkey4eva]

Are you a fan of adetailer's YOLOv8 face detection and/or the per-person matching logic?

Now you can use YOLOv8 in Swarm - https://github.com/Stability-AI/StableSwarmUI/blob/master/docs/Features/Prompt%20Syntax.md#automatic-segmentation-and-refining

tldr: download a yolov8 model like https://github.com/hben35096/assets/releases/download/yolo8/face_yolov8m-seg_60.pt to (Swarm)/Models/yolov8, then in swarm type a prompt like a photo of a woman standing in the street smiling <segment:yolo-face_yolov8m-seg_60.pt-1> a beautiful woman's smiling face with this new segment sub-syntax to automatically use YOLOv8 for the segmentation handling.

[mcmonkey4eva]

TensorRT support! Thanks to Comfy and nvidia for building the Comfy toolkit for TRT

Simply click the "Create TensorRT Engine" button in the menu for a model

you will then get a .engine file listed in the same folder (on filesystem this is actually in (Models)/tensorrt/)
TRT models are great for static servable models that run faster that normal models do (eg that you'll use over an API or just use frequently) but aren't compatible with all options (eg loras and wotnot are unsupported currently)

Note as well this is brand new / experimental and may have reliability issues

[mcmonkey4eva]

New Model Downloader Utility tab

Notably supports entering a civitai url and it will download the model + metadata for that model automatically

You can also click "Edit Metadata" on an existing model and give a civitai URL to load model from that url onto an existing model

[mcmonkey4eva]

As of 2024/06/21 StableSwarmUI will no longer be maintained under Stability AI.

The original developer will be maintaining an independent version of this project as mcmonkeyprojects/SwarmUI

Windows users can migrate to the new independent repo by simply updating and then running migrate-windows.bat

For Linux, Mac, or manual Windows: open a terminal in your Swarm folder then run git remote set-url origin https://github.com/mcmonkeyprojects/SwarmUI

See full migration guide here: mcmonkeyprojects/SwarmUI#2