NMKD Stable Diffusion GUI Changelog
What's new in NMKD Stable Diffusion GUI 1.11.0
Sep 6, 2023- New: LoRA and LyCORIS support
- New: LoRA training GUI (replaces Dreambooth, works with 8 GB GPUs)
- New: Option to auto-delete generated images afterwards (if they weren't copied to favs)
- Improved: Updater is much more stable now, and only re-installs updated/new packages
- Improved: External VAEs now also work with Diffusers models
- Improved: Image progress bar now takes high-res fix into account instead of running twice
- Improved: Max seed increased from 4294967295 to 9223372036854775807 (int64 max)
- Improved: VAE folders are now separate from model folders (again)
- Improved: Embeddings folder path can now be changed
- Improved: Progress bar and ETA now accounts for queued tasks instead of running per-task
- Improved: Image Viewer buttons (e.g. prev/next) are now disabled when there are no images to show
- Improved: Various UI/UX improvements
- Fixed issues caused by moving the installation folder, this should now work without problems
- Fixed: Applying CLIP Skip to a model would make it impossible to disable it again
- Fixed: Inpainting model name check is no longer case-sensitive and no longer requires a hyphen
- Fixed: "Append To Prompt" button for embeddings will no longer add a comma if there already is one
New in NMKD Stable Diffusion GUI 1.9.1 (Feb 1, 2023)
- New: Resolution can now also be adjusted in InstructPix2Pix mode (resizes input image)
- New: Resulting aspect ratio is shown next to resolution option
- New: Right-click image menu in InstructPix2Pix has an option to copy a comparison image
- Improved: Model compatibility, any Safetensors file should now work fine after conversion
- Improved: Model converter no longer includes 1 GB safety checker in ONNX outputs
- Improved: Image Viewer now shows image resolution info in InstructPix2Pix mode
- Improved: InstructPix2Pix no longer downloads unnecessary files, now <2GB
- Fixed: InstructPix2Pix errors when trying to run it without internet connection
- Fixed: Certain options result in filenames being too long causing issues
- Fixed: Stable Diffusion model gets reloaded if amount of free RAM changed
- Fixed: Image export breaks if initialization image(s) import takes too long
- Fixed: "Open Output Folder" opens Documents folder
- Fixed: ONNX and InstructPix2Pix don't accept images with UTF8 characters in path
- Fixed: Running InvokeAI post-processing sometimes fails due to long paths
- Fixed: Issues with comma vs dot, depending on Windows region settings
New in NMKD Stable Diffusion GUI 1.9.0 (Jan 30, 2023)
- New: Added InstructPix2Pix (Use with Settings -> Implementation -> InstructPix2Pix)
- New: Added the option to show the input image next to the output for comparisons
- New: Added option to choose output filename timestamp (None, Date, Date+Time, Epoch)
- Improved: minor UI fixes, e.g. no more scrollbar in main view if there is enough space
- Fixed: Dreambooth learning rate multiplier would default to 2x
- Fixed: Minor PNG metadata parsing issues
- Fixed: Various of other minor fixes