- Resolve GPU detection issues during installation.
- Added keyboard shortcuts for project view:
Ctrl+Up- selects previous outputCtrl+Down- selects next output
- Fixed issues with NVIDIA driver detection on Windows.
- Fixed issues with Intel Arc GPUs not being correctly utilized within PyTorch.
- Added project tagging to context menu and project list.
- Added a button to invert the mask in the mask editor.
- Added an option to enable experimental performance optimizations in backend settings.
- Improved design consistency of context menus.
- Made all project settings categories not collapsed by default.
- Added detection of custom installation paths when using the Windows installer.
- Added a check during installation to ensure application files aren't overwritten with custom paths.
- Resolved issues with installation task queue being stuck.
- Added an option to mark projects as favorite.
- Added experimental support for Intel Arc/Battlemage GPUs.
- Resolved issues with opening a large amount of projects at once.
- Resolved performance issues with loading the project list.
- Resolved issues with draft projects not being saved correctly.
- Added user accounts. (Web UI only.)
- Added keyboard shortcuts for projects:
Ctrl+N- opens a new projectCtrl+W- closes the current projectCtrl+Shift+W- closes the current project without savingCtrl+Left- switches to the previous project in the project listCtrl+Right- switches to the next project in the project list
- Added support for using Shift to skip the save dialog while closing draft projects.
- Added an option to close other tabs in the context menu.
- Fixed issues with project settings not being updated correctly.
- Fixed issues with optional features being broken after bundle reset.
- Fixed high CPU usage when downloading models.
- Added a "Temperature unit" switch to settings.
- Added a keyboard shortcut to reset weight of selected text in prompt.
- Added support for grouping prompt styles.
- Removed bundle reset actions for Docker images.
- Improved file update watcher performance.
- Improved prompt data import from image files.
- Updated Electron to v34.0.2.
- Improved application loading time.
- Fixed issues with log items not being refreshed in "System monitor".
- Fixed parenthesis insertion around words in prompts.
- Fixed metadata for downloaded models.
- Added a "Font size" slider to settings.
- Added a modal for model download confirmations and an input to select target folder.
- Added a close button to System Monitor.
- Improved design consistency.
- Fixed issues with popovers not showing up in the right place (e.g. model selection or mask image selection).
- Added a button to unload all models from VRAM.
- Added a new system monitor view with backend logs, loaded models and hardware utilization information.
- Model caching has been reworked to retain relevant models.
- Resolved problems with cleanup of draft projects.
- Restored missing context menu options - "Save Image As" and "Copy Image".
- Resolved an issue with the "Discard" button resulting in the old image being displayed as an output.
- Fixed an issue where some buttons near the top of the window weren't fully clickable.
- Added support for generating video using Hunyuan Video models.
- Added support for APNG outputs.
- Added an option to change the installation directory during setup.
- Added support for GGUF model files. (Requires the "GGUF" optional feature to be installed.)
- Added an option to change the default prompt.
- Added an option to use light theme.
- Added an option to always run VAE on CPU.
- Moved bundles to a faster CDN to improve installation speed.
- Improved word wrapping in prompt fields.
- Added the new file picker to path fields in settings.
- In CivitAI downloader, model rating was replaced by thumbs up/down count.
- The "Draft project" dialog is now only shown if the project has been changed.
- Fixed auto update.
- Fixed issues with GPU detection using
rocm-smion Linux.
Warning
Manual steps required to finish the update. If you have any CLIP or UNET models (in models/clip and models/unet), you will need to move them to models/text_encoder and models/diffusion_model respectively.
- Added a setting to reserve a certain amount of VRAM.
- Added support for more than 2 CLIP models in advanced mode.
- Added authorization for HuggingFace downloads.
- Added support for loading checkpoints that don't contain VAE/CLIP/UNET (e.g. SD 3.x) and then supplying external models.
- Added a button to reset selection in model fields.
- Added support for "metacheckpoints" - metadata files that combine multiple separate models into one virtual checkpoint.
- Added version to names of models downloaded from CivitAI.
- Added detection of GPU drivers during setup.
- Added support for DirectML on Intel Arc GPUs on Windows.
- Improved error messages during checkpoint loading (e.g. if a checkpoint with missing CLIP/VAE is used).
- ZLUDA is now automatically enabled during setup if compatible HIP SDK is installed.
- Added SD 3.x and FLUX.1 to downloadable models.
- Added TAESD support for SD 3.x and FLUX.1.
- Fixed ZLUDA support.
- Fixed notarization issues on macOS.
- Fixed setup error display (during extraction step).
- Fixed issues with saving/renaming projects on Windows.
- Fixed issues with inpainting masks not being correctly loaded.
- Fixed parenthesis logic in prompt text fields.
- Added experimental support for ZLUDA on Windows. Requires HIP SDK 5.7 to be installed, and a compatible GPU (or patches for older GPUs). Bundle reset required, "Use ZLUDA" needs to be selected in the Hardware section during setup.
- Fixed img2img errors.
- Added error handling for backend initialization.
- Fixed the app crashing on start.
- Added a button to easily postprocess (upscale) output images after generation.
- Added buttons to reset all settings (without resetting the bundle) and reset the bundle (without resetting settings) to "About Metastable".
- Added keyboard shortcuts (
Up/Down/Enter) for selecting models in model browser. - Added support for trackpad gestures in image previews.
- Added support for resizing the prompt area.
- Improved handling of window resizes in image previews.
- Updated ComfyUI code to latest version.
- Fixed display of input fields next to sliders.
- Fixed page zoom issues on macOS.
- Added support for adding extra model folders.
- Added a refresh button to model browser.
- Added storage of brush settings to mask editor.
- Resolved issues with AMD GPU detection on Windows.
- Fixed issues with mask editor reverting the mask at random.
- Fixed layout issues in project generation preview.
- Added an option to pad edges for inpainting.
- Added support for PuLID.
- Features are now split into modules that can be enabled/disabled.
- Moved backend status to settings.
- Reduced memory usage during image generation.
- Refactored model path handling.
- Resolved issues with non-square image generation previews.
- Resolved issues with GPU detection on Linux.
- Resolved issues with out-of-memory crashes.
- Fixed usage of TAESD decoder for previews.
- Fixed problems with project metadata sometimes not loading correctly.
- CivitAI tab now remembers search filters when switching between tabs.
- Keyboard shortcuts are now customizable.
- Prompt input has been updated to include new features:
- Added line numbers.
- Added keyboard shortcuts:
Ctrl+Up/Downto change the weight of certain parts of the prompt.Ctrl+/to comment out parts of the prompt.((Shift+9) to wrap parts of the prompt in parentheses after selecting something.
- Added support for comments:
- Inline/multiline:
#- comment -# - Line (Python style):
# comment
- Inline/multiline:
- Added syntax highlighting.
- Added support for fetching utilization data from AMD GPUs on Linux (via
rocm-smi).
- Resolved issues with detection of enterprise GPUs.
- Resolved "App quit unexpectedly." on macOS.
- Added an easy way to select input images for img2img/inpainting.
- Added confirmation before project settings are restored from an image.
- Added simplified sampling quality picker.
- Added simplified output image resolution picker.
- Added upload and management of input/mask files inside of projects.
- Input images are now automatically saved and can be easily reused.
- Project files are now automatically refreshed.
- Image thumbnails are now WEBP and support transparency.
- UI state (e.g. section collapsed state) is now properly saved.
- Error display now doesn't overflow the screen for long errors.
- Fixed issues with resetting the bundle.
- Added keyboard shortcuts:
Ctrl+Z/Ctrl+Shift+Zin mask editor now result in Undo/Redo respectively.Alt+Enterin project view now cancels the currently running prompt.
- Added Undo/Redo buttons to mask editor.
- Window position and state is now restored on launch.
- Added progress display in extraction task in the installer.
- Updated Electron to v32.1.2.
- Improved design consistency.
- Reduced application size.
- Resolved issues with bundle detection (should fix performance issues on AMD GPUs on Windows).
- Fixed CivitAI config file downloads.
- Added a back button to easily return from grid/editor.
- Added middle click to open project without selecting it.
- Added middle click to dismiss failed/queued tasks.
- Added a confirmation dialog when deleting multiple output files.
- Added support for loading YAML configuration files for checkpoints.
- Added an error handler for unrecoverable errors.
- Added storage of project settings section state (if the section is collapsed or not).
- Added a settings section for ComfyUI arguments.
- Improved the process of adding models to a project.
- Improved progress display on smaller screens.
- Fixed JSON serialization issues. (resolves problems with duplicating projects, etc.)
- Fixed display of large amounts of horizontal tabs.
- Updated design language.
- Replaced the output grid view with a more advanced file manager view.
- Added "Cancel" and "Clear queue" buttons to projects.
- Added a "Discard" button to quickly delete current output and try again.
- Added a context menu for text fields and images.
- Resolved issues with saving settings.
- Fixed copy/paste/select keyboard shortcuts on MacOS.
- Fixed CivitAI page navigation.
- Added prompt styles.
- Added a tag for corrupt files inside of the model browser.
- Added display of download errors in the download queue view.
- Added a list of recommended resolutions per model type and aspect ratio.
- Fixed crashes when corrupt model files are present.
- Fixed CivitAI authorization requirement detection for downloads.
- Added support for FLUX.1.
- Added architecture (e.g. SD1, SDXL) detection for models.
- Added display of error logs for failed generation tasks.
- Added sampling time statistics display.
- Added keyboard shortcuts to mask editor.
- Added menu to projects that allows the user to rename, delete and duplicate the project.
- Added a progress bar to app icon to display the current prompt progress.
- Added notifications to be displayed after prompt completion.
- Resolved an issue with AMD GPU detection.
- Fixed model metadata saving in downloads.
- Fixed usage of
Ctrl+Entershortcut in prompt fields.
- Support for Stable Diffusion 3 was added.
- Drag and drop is now supported in image input fields.
- In output image displays, a button to select the image as input in the current project was added.
- In output image displays, a button to delete the image was added.
- A button to reveal the model directory in file explorer was added.
- In inpainting mask editor, the right mouse button now erases the mask.
- User's Python site-packages do not conflict with the packages in the virtual environment anymore.
- ComfyUI's code was updated.
- Initial release.