v2.12.0

Added

New Packages

• Fooocus - mashb1t's 1-Up Edition by mashb1t
• Stable Diffusion WebUI reForge by Panchovix

Inference

• Added type-to-search for the Inference model selectors. Start typing while the dropdown is open to navigate the list.

  • 

• Added "Model Loader" option to Inference, for loading UNet/GGUF/NF4 models (e.g. Flux)

  • 

• Added support for the FP8 version of Flux in the default Model Loader via the "Use Flux Guidance" Sampler Addon

  • 

• Added trigger words to the Inference Extra Networks (Lora/Lyco) selector for quick copy & paste

• Image viewer context menus now have 2 options: Copy (Ctrl+C) which now always copies the image as a file, and Copy as Bitmap (Shift+Ctrl+C) (Available on Windows) which copies to the clipboard as native bitmap. This changes the previous single Copy button behavior that would first attempt a native bitmap copy on Windows when available, and fall back to a file copy if not.

• Added Face Detailer module to Inference

  • 
  • 

Package Manager

• Added Python dependencies override table to package installation options, where the default pip packages may be overriden for a package's install and updates. This can be changed later or added to existing packages through Package Menu > Python Dependencies Override

  • 
  • 

• Added "Change Version" option to the package card overflow menu, allowing you to downgrade or upgrade a package to a specific version or commit (#701, #857)

• Added "Disable Update Check" option to the package card overflow menu, allowing you to disable update checks for a specific package

• Added Custom commit option in the Advanced Options for package installs (#670, #839, #842)

• Added macOS support for Fooocus & related forks

• Added Intel OneAPI XPU backend (IPEX) option for SD.Next

Checkpoint Manager

• Added new Metadata Editor (accessible via the right-click menu), allowing you to create or edit metadata for models
  • 
• Added "New Directory" and "Delete" options to the context menu of the tree view.
• Added new toggle for drag & drop - when enabled, all selected models will now move together with the dragged model
• Added "File Size" sorting option
• Added "Hide Empty Categories" toggle
• Added "Select All" button to the InfoBar (shown when at least one model is selected)
• Added "Show NSFW Images" toggle

Model Browser

• Added "Hide Installed Models" toggle to the CivitAI Model Browser
• Added toggle to hide "Early Access" models in the CivitAI Model Browser
• Added ultralytics models to HuggingFace model browser

Other

• Added "Sign in with Google" option for connecting your Lykos Account on the Account Settings page
• Added zoom sliders for Outputs, Checkpoints, and Model Browser pages
• Added Settings option "Console: History Size" to adjust the number of lines stored in the console history when running packages. Defaults to 9001 lines.
  • 
• Added optional anonymous usage reporting for gauging popularity of package installs and features. You will be asked whether you want to enable this feature on launch, and can change your choice at any time in Settings > System > Analytics
• Added "Run Command" option in Settings for running a command with the embedded Python or Git executables
  • 
• Added "Enable Long Paths" option for Git to the Settings page
• Added "System Settings > Enable Long Paths" option to enable NTFS long paths on Windows
• Added Korean translations thanks to maakcode!
• (Windows, Linux) Added Vulkan rendering support using launch argument --vulkan. (On Windows, the default WinUI composition renderer is likely still preferrable. Linux users are encouraged to try the new renderer to see if it improves performance and responsiveness.)

Changed

• Optimized image loading across the app, with loading speed now up to 4x faster for local images, and up to 17x faster for remote images
• Image loading in the Outputs page now uses native memory management for ~2x less peak memory usage, and will release memory more quickly when switching away from the Outputs page or scrolling images out of view
• Improved animation fluidity of image rendering while scrolling quickly across large collections (e.g. Outputs, Model Browser)
• ComfyUI will no longer be pinned to torch 2.1.2 for nvidia users on Windows (#861)
• Model browser download progress no longer covers the entire card for the entire duration of the download
• Updated torch index to rocm6.1 for AMD users of ComfyUI
• Show better error message for early access model downloads
• Updated torch version for a1111 on mac
• Checkpoints tab now shows "image hidden" for images that are hidden by the NSFW filter
• OAuth-type connection errors in Account Settings now show a more detailed error message
• The "Download Failed" message for model downloads is now persistent until dismissed
• Separated the Generate button from the prompt control in Inference so it can be moved like other controls
• Updated translations for Turkish and Russian
• (Internal) Updated Avalonia to 11.1.3 - Includes major rendering and performance optimizations, animation refinements, improved IME / text selection, and improvements for window sizing / z-order / multi-monitor DPI scaling. (avaloniaui.net/blog/avalonia-11-1-a-quantum-leap-in-cross-platform-ui-development)
• (Internal) Updated SkiaSharp (Rendering Backend) to 3.0.0-preview.4.1, potentially fixes issues with window rendering artifacts on some machines.
• (Internal) Updated other dependencies for security and bug fixes.

Fixed

• Fixed #888 - error updating kohya_ss due to long paths
• Fixed some ScrollViewers changing scroll position when focus changes
• Fixed CivitAI Model Browser sometimes incorrectly showing "No models found" before toggling "Show NSFW" or "Hide Installed" filters
• Fixed SwarmUI settings being overwritten on launch
• Fixed issue where some Inference-generated images would be saved with the bottom missing
• Fixed #851 - missing fbgemm.dll errors when using latest torch with certain packages
• Fixed issue where ApproxVAE models would show up in the VAE folder
• Fixed #878 - Checkpoints tab will no longer try to browse directories it can't access
• Fixed crash when opening Settings page when refreshing CivitAI account status results in an error
• Fixed #814, #875 - Error when installing RuinedFooocus
• LORAs are now sorted by model name properly in the Extra Networks dropdown
• (macOS) Fixed OAuth connection prompts in Account Settings not automatically updating status after connection. Custom URL schemes are now also supported on macOS builds.

Supporters

Visionaries

• A heartfelt thank you to our Visionary-tier patron, Waterclouds! We greatly appreciate your continued support!

Pioneers

• A special shoutout to our Pioneer-tier patrons: tankfox, tanangular, Mr. Unknown, and Szir777! Your unwavering support means a great deal!

v2.11.8

Added

• Added Flux & AuraFlow types to CivitAI Browser
• Added unet folder links for ComfyUI thanks to jeremydk
• Added CLIP folder links for Forge

Changed

• Updated Brazilian Portuguese translations thanks to thiagojramos

Fixed

• Fixed #840 - CivitAI model browser not loading search results
• Fixed SwarmUI settings being overwritten on launch
• Fixed #832 #847 - Forge output folder links pointing to the incorrect folder
• Fixed errors when downloading models with invalid characters in the file name
• Fixed error when installing RuinedFooocus on nvidia GPUs

Supporters

Pioneers

• A big shoutout to our Pioneer-tier patrons: tankfox, tanangular, Mr. Unknown, and Szir777! We deeply appreciate your ongoing support!

v2.11.7

Changed

• Forge will use the recommended pytorch version 2.3.1 the next time it is updated
• InvokeAI users with AMD GPUs on Linux will be upgraded to the rocm5.6 version of pytorch the next time it is updated

Fixed

• Fixed Inference not connecting with "Could not connect to backend - JSON value could not be converted" error with API changes from newer ComfyUI versions

Supporters

Pioneers

• Shoutout to our Pioneer-tier supporters on Patreon: tankfox, tanangular, Mr. Unknown, and Szir777! Thanks for all of your continued support!

v2.11.6

Fixed

• Fixed incorrect IPAdapter download links in the HuggingFace model browser
• Fixed potential memory leak of transient controls (Inference Prompt and Output Image Viewer) not being garbage collected due to event subscriptions
• Fixed Batch Count seeds not being recorded properly in Inference projects and image metadata
• Fixed #795 - SwarmUI launch args not working properly
• Fixed #745 - not passing Environment Variables to SwarmUI

Supporters

Visionaries

• Shoutout to our Visionary-tier Patreon supporter, Scopp Mcdee! Huge thanks for your continued support!

Pioneers

• Many thanks to our Pioneer-tier supporters on Patreon: tankfox, tanangular, Mr. Unknown, and Szir777! Your continued support is greatly appreciated!

v2.11.5

Added

• Added DoRA category to CivitAI model browser

Fixed

• Fixed TaskCanceledException when adding CivitAI Api key or searching for models when the API takes too long to respond. Retry and timeout behavior has been improved.
• Fixed #782 - conflict error when launching new versions of Forge
• Fixed incorrect torch versions being installed for InvokeAI
• Fixed ArgumentOutOfRangeException with the Python Packages dialog ItemSourceView when interacting too quickly after loading.

Supporters

Visionaries

• Shoutout to our Visionary-tier Patreon supporters, Scopp Mcdee, Waterclouds, and our newest Visionary, Akiro_Senkai! Many thanks for your generous support!

Pioneers

• Many thanks to our Pioneer-tier supporters on Patreon, tankfox, tanangular, and our newest Pioneers, Mr. Unknown and Szir777! Your support is greatly appreciated!

v2.11.4

Changed

• Base Python install will now use setuptools==69.5.1 for compatibility with torchsde. Individual Packages can upgrade as required.
• Improved formatting of "Copy Details" action on the Unexpected Error dialog
• (Debug) Logging verbosity for classes can now be configured with environment variables (Logging__LogLevel__<TypeFullName>).

Fixed

• Fixed ComfyUI slower generation speed with new torch versions not including flash attention for windows, pinned torch==2.1.2 for ComfyUI on Windows CUDA
• Fixed #719 - Fix comments in Inference prompt not being ignored
• Fixed TaskCanceledException when Inference prompts finish before the delayed progress handler (250ms)

Supporters

Visionaries

• Huge thanks to our Visionary-tier supporters on Patreon, Scopp Mcdee and Waterclouds! Your support helps us continue to improve Stability Matrix!

Pioneers

• Thank you to our Pioneer-tier supporters on Patreon, tankfox and tanangular! Your support is greatly appreciated!

v2.11.3

Changed

• Base Python install will now use pip>=23.3.2,<24.1 for compatibility with torchsde.Individual Packages can upgrade as required.
• Added default PIP_DISABLE_PIP_VERSION_CHECK=1 environment variable to suppress notices about pip version checks.
  • As with other default environment variables, this can be overridden by setting your own value in Settings > Environment Variables [Edit].

Fixed

• Fooocus Package - Added pip>=23.3.2,<24.1 specifier before install, fixes potential install errors due to deprecated requirement spec used by torchsde.
• Fixed error when launching SwarmUI when installed to a path with spaces
• Fixed issue where model folders were being created too late in certain cases
• Fixed #683 - Model indexing causing LiteDB errors after upgrading from older versions due to updated enum values

Supporters

Visionaries

• Huge thanks to our Visionary-tier supporters on Patreon, Scopp Mcdee and Waterclouds! Your support helps us continue to improve Stability Matrix!

Pioneers

• Thank you to our Pioneer-tier supporters on Patreon, tankfox and tanangular! Your support is greatly appreciated!

v2.11.2

Changed

• StableSwarmUI installs will be migrated to SwarmUI by mcmonkeyprojects the next time the package is updated
  • Note: 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

Fixed

• Fixed #700 - cannot import 'packaging' error for Forge

Supporters

Visionaries

• Huge thanks to our Visionary-tier supporters on Patreon, Scopp Mcdee and Waterclouds! Your support helps us continue to improve Stability Matrix!

Pioneers

• Thank you to our Pioneer-tier supporters on Patreon, tankfox and tanangular! Your support is greatly appreciated!

v2.11.1

Added

• Added Rename option back to the Checkpoints page

Changed

• Unobserved Task Exceptions across the app will now show a toast notification to aid in debugging
• Updated SD.Next Package details and thumbnail - #697

Fixed

• Fixed #689 - New ComfyUI installs encountering launch error due to torch 2.0.0 update, added pinned numpy==1.26.4 to install and update.
• Fixed Inference image mask editor's 'Load Mask' not able to load image files
• Fixed Fooocus ControlNet default config shared folder mode not taking effect
• Fixed tkinter python libraries not working on macOS with 'Can't find a usable init.tcl' error

Supporters

Visionaries

• Shoutout to our Visionary-tier supporters on Patreon, Scopp Mcdee and Waterclouds! Your generous support is appreciated and helps us continue to make Stability Matrix better for everyone!

Pioneers

• A big thank you to our Pioneer-tier supporters on Patreon, tankfox and tanangular! Your support helps us continue to improve Stability Matrix!

v2.11.0

Added

Packages

• Added new package: SDFX by sdfxai
• Added ZLUDA option for SD.Next
• Added more launch options for Forge - #618
• Added search bar to the Python Packages dialog
• 

Inference

• Added Inpainting support for Image To Image projects using the new image mask canvas editor
• Added alternate Lora / LyCORIS drop-down model selection, can be toggled via the model settings button. Allows choosing both CLIP and Model Weights. The existing prompt-based <lora:model:1.0> method is still available.
• 
• Added optional Recycle Bin mode when deleting images in the Inference image browser, can be disabled in settings (Currently available on Windows and macOS)
• 

Model Browsers

• Added PixArt, SDXL Hyper, and SD3 options to the CivitAI Model Browser
• Added XL ControlNets section to HuggingFace model browser
• Added download speed indicator to model downloads in the Downloads tab

Output Browser

• Added support for indexing and displaying jpg/jpeg & gif images (in additional to png and webp / animated webp), with metadata parsing and search for compatible formats

Settings

• Added setting for locale specific or invariant number formatting
• Added setting for toggling model browser auto-search on load
• Added option in Settings to choose whether to Copy or Move files when dragging and dropping files into the Checkpoint Manager
• Added folder shortcuts in Settings for opening common app and system folders, such as Data Directory and Logs

Translations

• Added Brazilian Portuguese language option, thanks to jbostroski for the translation!

Changed

• Maximized state is now stored on exit and restored on launch
• Drag & drop imports now move files by default instead of copying
• Clicking outside the Select Model Version dialog will now close it
• Changed Package card buttons to better indicate that they are buttons
• Log file storage has been moved from %AppData%/StabilityMatrix to a subfolder: %AppData%/StabilityMatrix/Logs
• Archived log files now have an increased rolling limit of 9 files, from 2 files previously. Their file names will now be in the format app.{yyyy-MM-dd HH_mm_ss}.log. The current session log file remains named app.log.
• Updated image controls on Recommended Models dialog to match the rest of the app
• Improved app shutdown clean-up process reliability and speed
• Improved ProcessTracker speed and clean-up safety for faster subprocess and package launching performance
• Updated HuggingFace page so the command bar stays fixed at the top
• Revamped Checkpoints page now shows available model updates and has better drag & drop functionality
• Revamped file deletion confirmation dialog with affected file paths display and recycle bin / permanent delete options (Checkpoint and Output Browsers) (Currently available on Windows and macOS)

Fixed

• Fixed crash when parsing invalid generated images in Output Browser and Inference image viewer, errors will be logged instead and the image will be skipped
• Fixed missing progress text during package updates
• (Windows) Fixed "Open in Explorer" buttons across the app not opening the correct path on ReFS partitions
• (macOS, Linux) Fixed Subprocesses of packages sometimes not being closed when the app is closed
• Fixed Inference tabs sometimes not being restored from previous sessions
• Fixed multiple log files being archived in a single session, and losing some log entries
• Fixed error when installing certain packages with comments in the requirements file
• Fixed error when deleting Inference browser images in a nested project path with recycle bin mode
• Fixed extra text in positive prompt when loading image parameters in Inference with empty negative prompt value
• Fixed NullReferenceException that sometimes occurred when closing Inference tabs with images due to Avalonia.Bitmap.Size accessor issue
• Fixed #598 - program not exiting after printing help or version text
• Fixed #630 - InvokeAI update hangs forever waiting for input
• Fixed issue where the "installed" state on HuggingFace model browser was not always correct
• Fixed model folders not being created on startup

Supporters

Visionaries

• Shoutout to our Visionary-tier supporters on Patreon, Scopp Mcdee and Waterclouds! Your generous support is appreciated and helps us continue to make Stability Matrix better for everyone!

Pioneers

• A big thank you to our Pioneer-tier supporters on Patreon, tankfox and tanangular! Your support helps us continue to improve Stability Matrix!