SwarmUI - 0.9.8

It's been 6 months and almost 600 commits since 0.9.7... a lot has happened, here's just the most interesting things:

Major Updates

• Flux.2 support! https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Model%20Support.md#flux-2

  • Including Flux.2-Dev and Flux.2-Klein 4B and 9B, for both Text2Image and editing

• Z-Image support! https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Model%20Support.md#z-image

  • New installs by default include Z-Image Turbo now

• Anima model support! https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Model%20Support.md#anima

• Hunyuan Image 2.1 support! https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Model%20Support.md#hunyuan-image-21

• Hunyuan Video 1.5 Support! https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Video%20Model%20Support.md#hunyuan-video-15

• Qwen Image Edit Plus support! https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Model%20Support.md#qwen-image-edit

• LTX-2 support! https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Video%20Model%20Support.md#lightricks-ltx-video-2

  • First decent foss video model that can generate proper audio!

• Initial advanced "Auto-Scaling Backend" for cluster deployments -- still experimental, use at your own risk https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Features/AutoScalingBackend.md

• Comfy input/output folders now are forwarded to the Image History view

• Swarm has a patreon now! https://patreon.com/swarmui

• Audio and Video are now first-class citizens in SwarmUI. They have proper interfaces, full support in the history tab, and the ability to form input parameters that take them. This includes video files that contain audio.

• AMD on Windows now uses native ROCm-PyTorch-Windows!

• If you have accounts enabled, there's now a button to "Impersonate User", useful both to test the UI from other users' perspectives, and to set up alternate personal work environments

• Work on support for dedicated audio models has begun.

Minor Updates

• The new extension for SeedVR is getting a lot of love, it upscales your images a lot faster and smarter than most other methods

• Added the option to allow new users to register accounts on your swarm instance if you share it publicly (defaults disabled)

• added EasyCache params. EasyCache is a comfy built-in node that's like TeaCache and the other "step cache" techniques, but because it's native to comfy it just kinda works with everything instead of messy installs

• added qwen controlnet support

• added autorecovery for downloads failing during unstable internet, for install and model downloads

• trigger phrases now have a copy button

• improved the wildcard editing ui

• improved model class detection

• new TCFG param (thanks to @green-s )

• new "Catppuccin" themes (thanks to @mrblomblo )

• new stealth metadata settings (thanks to @Hugs288 )

• model pred type now has SD3 (RF) as an option because somebody made a hacky SDXL RF model

• modal image inputs (for editing models, wildcards, presets) are now nicer

• Chroma support improved, and Chroma Radiance support added

• Improved AuraFlow support to enable new Pony v7

• "MediaFile" rework to enable new types of data (replacing "Image")

• SwarmInputAudio and SwarmInputVideo nodes

• user setting ui.HideErrorMessages

• model metadata edits are now in-place, meaning they can often avoid rewriting the file to drive

• image editor now has a shape tool (to draw rectangle or circle shapes)

• installer now has a nicer theme selector

• model shortcodes in model selectors

• models can now link a Preset via the Edit Metadata menu

• Made a megagrid of parameters for Z-Image Turbo https://sd.mcmonkey.org/zimagegrid/#auto-loc,true,true,false,true,false,cfgscale,steps,none,none,extremecloseupt,4,1,3,1024x1024,1,euler,simple

• Support for Ovis 7B but the model kinda sucks https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Model%20Support.md#ovis

• The comfy multigpu input now has "Use Swarm Queue" as an option

• Internal reorganizations to prepare for native LLM support

• you can now star presets (thanks to @hansneddyanto )

• new user setting AudioBehavior

• new server setting UseSecondaryTriggerPhraseSources to let you turn off the tag spam some models tend to have

• Extensions can now have their own csproj files, see here for detail https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/Making%20Extensions.md

• as usual, many small improvements, bug fixes, docs improvements, etc. I probably forgot at least a few worth adding here, the change list is loooooong.