Reddit thread for this release: https://www.reddit.com/r/StableDiffusion/comments/1jztcuu/swarmui_096_release/

Major Updates

• Added support for Wan 2.1 video models https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Video%20Model%20Support.md#wan-21
• Added support for Lumina 2 image model https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Model%20Support.md#lumina-2
• Added support for SkyReels video model https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Video%20Model%20Support.md#skyreels-text2video
• Added support for Hunyuan Image2Video model and its "v2/fixed" variant https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Video%20Model%20Support.md#hunyuan-video
• Prompt syntax upgrades:
  • new <extend:...> syntax to automatically extend videos using an image2video model
  • new <refiner> syntax to add a custom prompt or lora to the refine/upscale section
  • <region:...> blocks can now contain <lora:...>s
• Documented how to use Triton, Torch.Compile, SageAttention on Windows: https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Advanced%20Usage.md#triton-torchcompile-sageattention-on-windows
• Swarm has a linkable website now https://swarmui.net/
• Multi-User account system preview https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Sharing%20Your%20Swarm.md
  • Have programming / cybersec experience? Check the public call for security researchers #679
• Nunchaku SVDQuant support https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Model%20Support.md#nunchaku-mit-han-lab

Minor Updates

• The old manual update bat/sh scripts are removed, in favor of using the UI button in the Server tab, or the auto-update option in server config. If you love script files you can always just slap in your own git pull script
• Comfy node autoupdates are now managed by a separate backend setting for ComfyUI autoupdates
• New installs of Swarm will use Python 3.12 on the backend by default now
• yolo segments now support class ids, and a SegmentSortOrder param
• added Renorm CFG under advanced sampling, for Lumina 2
• new Color Correction Behavior advanced parameter to try to improve flux inpainting, with thanks to @willhsmit for helping with implementation
• AccessControlAllowOrigin server setting
• initial handling for RTX 50 series, full support pending pytorch general availability support ofc
• Video feature updates: new gif-hd and h265-mp4 export formats, animated webps can now be used in Init Image to do batch or t2v gens, frameinterp params for t2v, added GIMM-VFI support, and several other minor improvements to enhance how well Swarm works around videos
• Video2Video Creativity param under image to video so you can refine video directly with an i2v model
• Trim Video Start/End Frames parameters to cut the edge frames than tend to bug on Wan and the initial Hunyuan I2V model
• new server config setting instance title, so for shared instances or if you have multiple instances, you can label it in the title and welcome message
• new setting setting DownloadToRootID to have multiple model folders and select which one to download to, docs here https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Advanced%20Usage.md#multiple-model-folders
• new param Use CFG Zero Star to use this https://arxiv.org/abs/2503.18886
• new internal param DoNotSaveIntermediates for if you want to output intermediates but not save them
• new server setting: ModelLoadOrderPreference for when you have multiple backends, to adjust preferences for which backends get used
• Swarm-API-Backend now has a ConnectionAttemptTimeoutSeconds option
• new user setting EnterKeyGenerates so you can disable hitting enter in a prompt box to generate
• new Flux Disable Guidance param for flux-dev based models that don't want flux guidance (eg flex.1?)
• Very large model lists are now much more responsive
• new user setting ImageMetadataFormat to configure how image metadata displays. Also the sidebar metadata display is in table format now
• as usual, many small improvements, bug fixes, docs improvements, etc.