v2.14.0

Hey everyone, the wait is over! Stability Matrix v2.14.0 has officially landed! ✨

After a packed development cycle (thanks for your patience!), we're thrilled to roll out a massive update with new features, integrations, performance boosts, and important fixes based on your feedback.

🪄 Introducing: Prompt Amplifier

Need inspiration or want to add more depth to your prompts? Click the magic wand 🪄 in the Inference prompt editor to use the Prompt Amplifier! Our AI assistant, powered by our experimental Spark model, will then generate creative expansions of your prompt.

⚡ Accelerated Model Discovery (Beta for Insider+)

Tired of waiting for search results? We hear you! Accelerated Model Discovery utilizes our optimized server connections to dramatically speed up browsing online repositories like Civitai. This optional connected enhancement was made possible by community support – local features will always remain free and open-source.

model_discovery.mp4

✨ More v2.14.0 Highlights: ✨

• New Packages: Welcome Stable Diffusion WebUI AMDGPU Forge and Forge - Classic! Plus, easily install Triton & SageAttention for ComfyUI.

• Enhanced Inference: Added HiDream and Wan 2.1 T2V/T2I video projects support, base model filtering for Extra Networks, Rescale CFG, Align Your Steps scheduler, wildcard support ({a|b|c}), Plasma Noise, and a handy width/height swap button.

• 

• Smarter Prompt Editing: AST parsing improves prompt understanding, plus Ctrl+Up/Down hotkeys to easily adjust token weights!

• Expanded Model Repos: Added an OpenModelDB tab and Wan 2.1 files to the HuggingFace browser.

• Compatibility & Bug Fixes: Upgraded HIP SDK, fixed issues with RTX 5000, macOS file handling & display quirks, package updates, model metadata, image caching, and much more!

A massive thank you to our incredible Patreon Supporters! Your support fuels the continued development and maintenance of Stability Matrix. We couldn't do this without you!

Full Changelog

Added

New Packages & Integrations

• Added new package - Stable Diffusion WebUI AMDGPU Forge
• Added new package - Stable Diffusion WebUI Forge - Classic
• Added new Package Command (in the 3-dots menu) for installing Triton & SageAttention in ComfyUI

Inference Features

• Added Prompt Amplifier to Inference - click the magic wand 🪄 in the prompt editor to expand and enrich your ideas. Tailor the vibe with the ‘Feel’ selector and watch as your generations come to life with extra detail, coherence, and flair!
• Added support for HiDream in Inference - see ComfyUI Examples for more details
• Added toggle for filtering Inference Extra Networks by base model
• Added Rescale CFG addon to Inference
• Added Swap Dimensions button between the width/height input in Inference
• Added Ctrl+Tab/Ctrl+Shift+Tab shortcuts for navigating between Inference tabs
• Added Align Your Steps scheduler to Inference
• Added wildcards to Inference prompts, e.g. {blue|green|red} will randomly select one of the colors
• Added Wan 2.1 Text to Video and Text to Image project types for Inference
• Added new autocomplete tag source to Inference - Danbooru/e621 merged tags
• Added Abstract Syntax Tree (AST) parsing for Inference prompts. This provides a more robust internal understanding of prompt structure, paving the way for future enhancements.
• Added hotkey (Ctrl+Up/Ctrl+Down) in Inference prompt editors to adjust the weight emphasis of the token under the caret or the currently selected text.
  • This automatically wraps the token/selection in parentheses () if it's not already weighted.
  • It modifies existing weights within parentheses or adds weights if none exist (e.g. (word:1.1)).
  • Handles selection spanning multiple tokens intelligently.
• Added Plasma Noise addon to Inference for text to image workflows

Model Management & Discovery

• Added Accelerated Model Discovery (Beta) (⚡ icon in Civitai Browser) for Insider+ supporters. Utilizes an optimized connection for dramatically faster, more responsive browsing of online model repositories.
• Added OpenModelDB tab to the Model Browser
• Added Wan 2.1 files to the HuggingFace model browser

User Interface & Experience (UI/UX)

• Added Undo/Redo commands to text editor context menus

Internal / Developer Changes

• (Internal) Introduced unified strategy pattern (IConfigSharingStrategy) to for handling different config file formats (JSON, YAML, FDS).
  • Added support for configuring nested paths in JSON and YAML files (e.g. paths.models.vae) via dot-notation in SharedFolderLayoutRule.ConfigDocumentPaths.
  • Packages can now use the SharedFolderLayout property to define a ConfigFileType and ConfigSharingOptions (like RootKey), without needing to implement custom configuration logic.

Changed

Inference Features

• Improved the quality of Inference inpainting by upgrading the workflow behind the scenes. The workflow remains the same for you — just better results!
• FaceDetailers in Inference will now inherit the primary sampler/scheduler/etc. by default. You can still manually set these by enabling the options via the ⚙️ button on the FaceDetailer card
• Slightly rearranged the FaceDetailer card layout due to the above change
• Inference "Extra Networks" selector now filters extra networks based on the selected base model
• Merged Inference GGUF workflows into the UNet model loader option (no longer need to choose GGUF separately)

Model Management & Discovery

• Changed the names of some of the shared model folders to better reflect their contents
• Improved Checkpoint Manager memory usage (thanks to @firegeek for the profiling assistance!)
• Performance optimizations for Checkpoint Manager (progress indicators now fully uses Compiled Bindings)

Package Management & Compatibility

• Upgraded HIP SDK installs to 6.2.4 for ComfyUI-Zluda and AMDGPU-Forge
• Updated install for kohya_ss to support RTX 5000-series GPUs

User Interface & Experience (UI/UX)

• Improved window state handling
• Updated some date strings to take into account the user's locale

Localization

• Updated Japanese, Brazilian Portuguese, Chinese, and Russian translations

Internal / Developer Changes

• (Internal) Upgraded FluentAvalonia to 2.3.0
• (Internal) Refactored configuration-based shared folder logic: Centralized handling into SharedFoldersConfigHelper and format-specific strategies, removing custom file I/O logic from individual package classes for improved consistency and maintainability.
  • Migrated packages ComfyUI (incl. Zluda), VladAutomatic (SD.Next), Sdfx, and StableSwarm to use the unified system for configuration and symlink based sharing.

Fixed

Installation, Compatibility & Core Functionality

• Fixed RTX 5000-series GPU detection in certain cases
• Fixed Package Updates and Change Version not using stored PyTorch index and instead using the default recommended index.
• Fixed ComfyUI-Zluda not being recognized as an option for Inference or SwarmUI (for real this time)
• Fixed errors from invalid pip specifiers in requirements files

User Interface & Experience (UI/UX)

• Fixed Image Viewer animation loader keeping file handles open, which resolves 2 different issues (OSes are fun):
  • (macOS) Fixed FileNotFoundException crash when navigating to Inference tab after deleting a Webp animation file previously opened in the Image Viewer Dialog.
  • (Windows) Fixed Webp animation files unable to be deleted without closing the app first.
• Fixed Image Viewer FileNotFoundException on fetching image size, if navigating to a deleted image file.
• (macOS) Fixed Webp / Gif animations RGB colors flipped.
• Fixed window disappearing on macOS when the saved window size is very small
• Fixed large white boxes appearing when tooltips are visible on macOS/Linux
• Fixed package images sometimes showing as blank due to concurrent image caching. Requests to same image resources are now de-duplicated
• Reduced memory usage from ShowDisabledTooltipExtension

Inference & Workflows

• Fixed some cases of missing custom nodes in SwarmUI
• Fixed Inference ControlNet Preprocessors using incorrect resolution and increased maximum of smallest dimension to 16384
• Fixed Inference Extra Networks card not allowing for more than one model at a time

Model Management & Discovery

• Fixed missing base model options in the Metadata Editor
• Fixed some crashes when using Accelerated Model Discovery

Supporters

Visionaries

Our heartfelt gratitude goes out to our amazing Visionary-tier Patrons: Waterclouds, Corey T, bluepopsicle, Bob S, Akiro_Senkai, and Ibixat! Your exceptional support is fundamental to the ongoing development and success of Stability Matrix. We are immensely grateful for your partnership and belief in the project! 🙏

Pioneers

We also want to give a huge thank you to our dedicated Pioneer-tier Patrons: tankfox, Mr. Unknown, Szir777, Tigon, NowFallenAngel, Al Gorithm, and welcom...

v2.13.4

Added

• Added support for RTX 5000-series GPUs in ComfyUI, Forge, and reForge
• Added "Rebuild .NET Project" command to SwarmUI installs - available via the 3-dots menu -> Package Commands -> Rebuild .NET Project

Changed

• Upgraded ComfyUI CUDA torch to 12.6
• Upgraded Lykos account connection to use OAuth 2.0 device flow
• (Internal) Updated Avalonia to 11.2.5

Fixed

• Fixed #1128 - overwriting models when downloading multiple with the same name
• Fixed ROCm torch indexes for ComfyUI & Forge
• Fixed model browser sometimes downloading to ModelsLora or ModelsStableDiffusion folders instead of the correct folder
• Fixed incorrect Unet folder path for ComfyUI users on Linux/macOS
• Fixed #1157 - crash when broken symlinks exist in model directories
• Fixed #1154 - increased width for package name on the package cards
• Fixed ComfyUI-Zluda not being recognized as an option for Inference or SwarmUI
• Fixed SwarmUI showing Python options in the 3-dots menu
• Fixed SD.Next install failures in certain cases when using Zluda

Supporters

Visionaries

• Huge thanks to our amazing Visionary-tier Patrons, Waterclouds and Corey T! We're truly grateful for your continued generosity and support!

Pioneers

• Special appreciation to our fantastic Pioneer-tier Patrons: tankfox, Mr. Unknown, Szir777, Tigon, NowFallenAngel, and our newest addition, Al Gorithm! Thank you all for your incredible commitment and ongoing encouragement!

v2.13.3

Added

• Added Safetensor Metadata viewer to the Checkpoint Manager context menu - thanks to @Genteure!

Changed

• "Remove symbolic links on shutdown" option now also removes links from Output Sharing

Fixed

• Fixed #1083 - "Show Nested Models" incorrectly displaying models from some non-nested folders
• Fixed #1120 - crash when right clicking in the console after restarting a package
• Fixed issue with InvokeAI model sharing when the host address is set to 0.0.0.0
• Fixed issue when parsing index URLs in Python Dependencies Override menu
• Fixed issue where models were filtered incorrectly in the Checkpoint Manager
• Fixed ComfyUI-Zluda not using the user-defined pip overrides

Supporters

Visionaries

• A heartfelt thank you to our incredible Visionary-tier Patrons, Waterclouds, TheTekknician, and Corey! Your unwavering support means the world to us!

Pioneers

• A big shoutout to our outstanding Pioneer-tier Patrons, tankfox, Mr. Unknown, Szir777, and NowFallenAngel! We deeply appreciate your ongoing support and dedication!

v2.13.2

Changed

• Suspended new SimpleSDXL installs due to security concerns. Existing installs will show a warning before being able to launch. For more information see GHSA-qq8j-phpf-c63j. (Thanks to @iwr-redmond for the detailed report.)

Supporters

Visionaries

• Many thanks to our amazing Visionary-tier Patrons, Waterclouds and TheTekknician! Your support is greatly appreciated!

Pioneers

• Shoutout to our Pioneer-tier Patrons, tankfox, Mr. Unknown, Szir777, Tigon, and NowFallenAngel! Thank you for your continued support!

v2.13.1

Changed

• Redesigned the Checkpoint Manager Filter flyout to include more options and improve the layout
• "Clear All" button will now remain at the top of the Downloads list regardless of scroll position - thanks to @Genteure!
• Improved image metadata parsing - thanks to @Genteure!

Fixed

• Fixed #1078 - "Call from invalid thread" error after one-click install finishes
• Fixed #1080 - Some models not displayed in Checkpoint Manager
• Fixed Inference image selector card buttons taking up the whole height of the card
• Fixed Inference mask editor failing to paint to the right-most edge on large images
• Fixed Inference mask editor not showing the entire image in certain circumstances
• Fixed crash when dragging & dropping images in Inference (hopefully)
• Fixed an issue where certain sampler/scheduler combos would not get saved in image metadata - thanks to @yansigit!

Supporters

Visionaries

• A heartfelt thank you to our exceptional Visionary-tier Patreon backers, Waterclouds and TheTekknician! We truly appreciate your steadfast support!

Pioneers

• We are also very grateful to our wonderful Pioneer-tier Patreon supporters, tankfox, Mr Unknown, Szir777, Tigon, and NowFallenAngel! Your support means a lot to us!

v2.13.0

Added

• Added new package - ComfyUI-Zluda - for AMD GPU users on Windows
• Added file sizes to the Checkpoint Manager tab
• Added the Discrete Model Sampling addon for Inference samplers, allows selecting different sampling methods, such as v_prediction, lcm, or x0, and optionally adjusts the model’s noise reduction strategy with the zero-shot noise ratio (ZSNR) toggle.
• Added Default GPU override in Settings -> System Settings -> Default GPU
• Added new "Copy" menu to the Inference gallery context menu, allowing you to copy generation parameters as well as the image
• Added "StableDiffusion" folder as an option when downloading Flux models in the CivitAI model browser
• Added support for SD3.5 in Inference
• Added CLIP_G to HuggingFace model browser
• Added search bar to the Installed Workflows tab
• Added "Search with Google" and "Search with ChatGPT" to the package console output & install progress console output context menus
• Added "Date Created" and "Date Last Modified" sorting options to the Checkpoints tab
• Added a new "Extension Packs" section to the extension manager, allowing you to create packs for easier installation of multiple extensions at once
• Added "Search by Creator" command to Civitai browser context menu
• Added Beta scheduler to the scheduler selector in Inference
• Added zipping of log files and "Show Log in Explorer" button on exceptions dialog for easier support
• Added max concurrent downloads option & download queueing for most downloads
• Added the ability to change the Models directory separately from the rest of the Data directory. This can be set in Settings > Select new Models Folder
• Added InvokeAI model sharing option

Changed

• Improved Packages Page grid layout to dynamically stretch to fill available space
• Text Encoder / CLIP selection in Inference is now enabled via the cogwheel ⚙️ button next to the model selector
• Updated Civitai model descriptions to properly render the interactive elements
• Adjusted the Branch/Release toggle during package install flow to be a little more obvious
• Updated the Dock library used for Inference - fixes some weirdness with resizing / rearranging panels
• New file format and key derivation for protecting locally encrypted secrets (i.e. Civit / Lykos accounts) that is no longer dependent on the OS Version. This should prevent system updates from clearing account logins.
• (Internal) Updated to .NET 9 Runtime and Avalonia 11.2.2 for performance improvements, lower memory usage, and bug fixes
• Updated pytorch index to rocm6.2 for AMD users of ComfyUI on Linux

Fixed

• Fixed text alignment issues in the Downloads tab for certain long names / progress infos
• Improved startup performance and resource usage with optimizations to hardware lookups. Moved reflection usages in dependency injection to source generation.
• Fixed an issue with ComfyUI-Impact-Subpack not being installed when using FaceDetailer in Inference
• Fixed GGUF models not showing in Inference without the GGUF extension installed (this means it will now properly prompt you to install the extension as well)

Supporters

Visionaries

• We're extremely grateful to our incredible Visionary-tier Patreon supporters, Waterclouds and TheTekknician! Thank you very much for your unwavering support!

Pioneers

• Many thanks to our amazing Pioneer-tier Patreon supporters, tankfox, Mr Unknown, Szir777, and our newest Pioneer, NowFallenAngel! Your generous support is very much appreciated!

v2.12.4

Added

• Added new package - CogVideo - many thanks to @NullDev for the contribution!
• Added more formatting options for Inference output filenames - thanks to @yansigit!

Changed

• Model browser base model types are now loaded dynamically from CivitAI, reducing the need for updates to add new types

Fixed

• Fixed crash when clicking "Remind me Later" on the update dialog
• Fixed some cases of crashing when GitHub API rate limits are exceeded
• Fixed Git missing from env vars when running SwarmUI
• Fixed missing package thumbnails due to moved or inaccessible urls
• Fixed an issue with updating FluxGym in certain cases - thanks to @NullDev!
• Fixed a typo in the Japanese translation - thanks to @mattyatea!

Supporters

Visionaries

• A huge thank you to our dedicated Visionary-tier Patreon supporter, Waterclouds! We’re thrilled to have your ongoing support!

Pioneers

• Shoutout to our great Pioneer-tier patrons: tankfox, tanangular, Mr. Unknown, Szir777, and our newest Pioneer, Tigon!. Your continued support is greatly appreciated!

v2.12.3

Added

• Added new package - SimpleSDXL - many thanks to @NullDev for the contribution!
• Added new package - FluxGym - many thanks to @NullDev for the contribution!
• Added more base model types to the CivitAI Model Browser & Checkpoint Manager

Changed

• Updated Russian translations thanks to @vanja-san
• Updated Simplified Chinese translations thanks to @QL-boy

Fixed

• Fixed some cases of FileTransferExists error when running re/Forge or Automatic1111
• Fixed update check not happening on startup for some users
• Fixed error when installing Automatic1111 on macOS
• Fixed some instances of Civitai model browser not loading new results
• Fixed "Unsupported Torch Version: Cuda" errors when installing a1111

Supporters

Visionaries

• A huge shout-out to our dedicated Visionary-tier Patreon supporter, Waterclouds! Your unwavering support is greatly appreciated!

Pioneers

• We'd also like to express our gratitude to our amazing Pioneer-tier patrons: tankfox, tanangular, Mr. Unknown, and Szir777! Your ongoing support means a great deal!

v2.12.2

Added

• Added Beta scheduler to the scheduler selector in Inference

Changed

• (Internal) Updated to Avalonia 11.1.4

Fixed

• Fixed ComfyUI NF4 extension not installing properly when prompted in Inference
• Fixed #932, #935, #939 - InvokeAI failing to update
• Fixed repeated nested folders being created in Models/StableDiffusion when using Forge in Symlink mode in certain conditions. Existing folders will be repaired to their original structure on launch.
• Fixed minimize button not working on macOS

Supporters

Visionaries

• We extend our heartfelt appreciation to our dedicated Visionary-tier Patreon supporter, Waterclouds. Your ongoing support is invaluable!

Pioneers

• We’d also like to thank our great Pioneer-tier patrons: tankfox, tanangular, Mr. Unknown, and Szir777. Your continuous support means a lot!

v2.12.1

Fixed

• Fixed #916 - InvokeAI failing to install/update on macOS
• Fixed #914 - Unable to use escaped colon : character in Inference prompts
• Fixed #908 - Forge unable to use models from "unet" shared folder
• Fixed #902 - Images from shared outputs folder not displaying properly in Stable Diffusion WebUI-UX
• Fixed #898 - Incorrect launch options for RuinedFooocus
• Fixed index url parsing in Python Packages window causing some packages to not have versions available
• Fixed a crash when switching between Model Sharing options for certain packages

Supporters

Visionaries

• A sincere thank you to our valued Visionary-tier Patreon supporter, Waterclouds. Your continued support is truly appreciated, and we’re grateful to have you with us on this journey.

Pioneers

• We’d also like to extend our gratitude to our Pioneer-tier patrons: tankfox, tanangular, Mr. Unknown, and Szir777. Your ongoing support means a great deal to us!