Reddit thread about this https://new.reddit.com/r/StableDiffusion/comments/1h81y4c/swarmui_094_release/
Major Updates
- added support for alimama flux inpaint controlnet - see notes at #1 (comment)
- added support for BFL's Flux Tools (redux, canny, depth, fill) - see notes at https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Model%20Support.md#flux1-tools
- added support for Genmo Mochi 1 Text-To-Video - see notes at #1 (comment)
- SD 3.5 Large & Large Turbo, and Medium are supported, including the Large controlnets - see notes at https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Model%20Support.md#stable-diffusion-35-large
- Lightricks LTX-Video text2video & image2video support - see notes at https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Model%20Support.md#lightricks-ltx-video
- UI feature: 'prompt add button' to make it easy to use advanced prompt syntax like Segment or Region - #1 (comment)
Minor Updates
- there's now a "popover on hover with delay" option in user settings if don't like either the
?
popups or the popover on hover instantly - fleshed out the troubleshooting doc https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Troubleshooting.md
- added better special error detection & handling (eg Swarm can now tell you clearly in logs if a crash happened due to RAM issues and how to fix it if so)
- automatic model downloading is now improved, and can even autodownload Flux/SD35 VAE if needed and all
- new parameter
VAE Tile Overlap
lets you control Tiled VAE quality - improved image metadata format and documented it https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Image%20Metadata%20Format.md
- there's now a server option to enable model hashes in image metadata
- there's now a
Copy Raw Metadata
button on images - the
Prompt Syntax
doc now has pictures throughout to demo the function https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Features/Prompt%20Syntax.md - the
Model Support
doc now has pictures to compare the models with https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Model%20Support.md - there's now a nicer batch view while generating, with model icon placeholders, blur effect, loading spinner, etc. (you can toggle the load spinner via the gear menu above the batch view)
- added advanced parameter
Segment Mask Oversize
to allow even finer control of segmentation behavior - new
MaxPromptLines
user setting to configure the visual scroll limit of the prompt box - preset select now persists across page reloads along with other parameters
- can now uninstall extensions
- can now use nonstandard IPAdapter models by saving them to the ipadapter models folder
- as usual, many small improvements, bug fixes, etc.
Security Notice: Ultralytics Breach 2024-12-05
On 2024-12-05, for a period of several hours, a version of the "Ultralytics" python package was marked latest on pip that contained crypto-miner malware. Swarm uses this package for "YOLO" model autosegmentation features. This was marked as version 8.3.41
and 42
If the following describes you:
- You are running Linux or Mac (not windows)
- you installed Swarm on 2024-12-05, or manually updated python packages that day (Swarm does not autoupdate these normally)
- (Possibly also only if) You used YOLO models on that day (Swarm installs the package, but doesn't load or run it until you actually try to use a yolo model)
Then you may be infected with a crypto miner. This will be visible as high CPU usage from a process named ultralytics
visible in top
. Killing the process and deleting the executable may suffice to end its control, though manage your personal system security to your own standards and at your own risk, it's hard to offer any guarantees about whether or not they slipped anything else in.
You can read more about this attack in this thread: ultralytics/ultralytics#18027
Swarm will not autoinstall those versions anymore, and also will remove existing installs of it and give a warning in your console.