forked from n00mkrad/text2image-gui
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathchangelog-motd.txt
145 lines (127 loc) · 9.17 KB
/
changelog-motd.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
SD GUI 1.8.1 Changelog:
- New: RunwayML inpainting now also works with ONNX implementation
- New: Version now checks for the latest version on startup (no updater yet)
- Improved: High-Res Fix can now be enabled if any axis is >512px instead of both
- Improved: High-Res Fix option is saved/loaded when closing and re-opening GUI
- Improved: Inpainting Mask Blur is now automatically disabled when using RunwayML inpainting
- Improved: In Installer, a custom git commit can now be used (for developers)
- Fixed: Upscalers were disabled by default on <=6GB GPUs
- Fixed: Model loading would fail without an internet connection
- Fixed: ONNX seeding did not work
- Fixed: CFG Scale <=1 didn't work or would fallback to default value
- Fixed: Inpainting mask was saved with irreversible blur, making editing harder
- Fixed: Init image import would ignore stretch/pad setting
- Fixed: Saved inpainting mask wouldn't get cleared properly when resetting it
- Fixed: Various minor bugs
SD GUI 1.8.0 Changelog:
- New: Stable Diffusion DirectML implementation, enables image generation on AMD GPUs
- New: Inpainting models are now supported, providing much better quality than the old method
- New: Models are cached in RAM when switching. Switching back loads a model in ~2 seconds
- New: Text-based masking - Describe what you want to mask instead of drawing the mask
- New: Initialization image can be previewed by Shift+Hovering over the filename in the GUI
- New: Generation resolution is automatically set to init image size if it doesn't match
- New: Pop-Up Image Viewer now has an option to enable Slideshow Mode by default
- New: Model converter (Dev Tools) to convert between Pytorch, Diffusers, ONNX, Safetensors
- New: Image Import window has a new option to use metadata plus the image itself as init image
- Improved: Image Viewer now has hotkeys Ctrl+Left/Ctrl+Right for previous/next image
- Improved: Image Viewer hotkeys (copy, open, etc.) now also work in pop-up viewer
- Improved: Mask Editor can be resized, is no longer locked to the init image size
- Improved: Mask Editor now has paste, invert, save, load, and undo functionality
- Improved: Inpainting mask can now be edited without having to clear it first
- Improved: Prompt History will not save a new entry if it's identical to the previous one
- Improved: Prompt History now has a limit of 500 entries to avoid long loading times
- Improved: Prompt Queue now behaves like a queue, not like a stack (FIFO)
- Improved: Welcome Screen (MOTD) can now be disabled after it's been shown once per version
- Improved: Image Import window now shows resolution, import actions are a dropdown to avoid clutter
- Improved: Mask Editor performance is better now, blur is only applied after releasing mouse
- Improved: User can now ignore the Dreambooth VRAM warning
- Fixed: Post-processing after generation would cause a soft-lock
- Fixed: Loading settings from prompt history sets incorrect values
- Fixed: Broken or malicious model files would cause an unhandled exception
- Fixed: Using a concept with <Name> syntax would show warning about missing placeholder
- Fixed: InvokeAI CLI wouldn't open if GUI image generation was cancelled before
- Fixed: Unhandled error when trying to use init images that no longer exist
- Fixed: GUI would freeze if no model files exist in models folder(s)
SD GUI 1.7.2 Changelog:
- Improved prompt syntax converter (converts A1111 or pre-1.7 weighting to new syntax)
- VAE Model selection now also lists .pt files (same format as ckpt, just different extension)
- Fixed init images being in random order after importing
- Fixed idle CPU usage
- Fixed an issue with having the same model filename in multiple model folders
- Fixed issue where k_diffusion module was not imported properly
- If a config YAML exists next to a model file it will be used instead of the default (e.g. model.ckpt.yaml)
SD GUI 1.7.1 Changelog:
- Added DPM++2 and DPM++2A samplers
- Added advanced syntax for step count (run same image with different step counts)
- New prompt attention syntax - Use (text)++ instead of ((text)), etc.
- Added automatic conversion of old attention syntax to new syntax
- Security: Added pickle malware scanner (checks model files automatically)
- Minor UI improvements, DreamBooth now has a dedicated button
- Fixed issues with pytorch not working correctly (No CUDA GPU detected)
- Fixed issues with post-processing after generation
- Fixed negative prompt input being visible when using Low Memory Mode
- Fixed certain sliders loading invalid values after restarting the program
- Reduced installation size and optimized code
SD GUI 1.7.0 Changelog:
- Added support for running multiple init images at once, e.g. for animation frames
- Added support for wildcards (insert words/phrases into prompt dynamically)
- Added "Hi-res Fix" which allows better results at higher (>512px) resolutions
- Added support for loading custom VAE models (can improve image quality)
- Added support for running upscaling or face restoration manually
- Added support for all samplers when using an init image (not just DDIM)
- Added button to resize the window to fit the currently displayed image
- Added option to save output images in a subfolder per session
- Added options to use seamless mode only on one axis (horizontal or vertical)
- Added a label that displays the current image's prompt
- Added hotkeys: Quick-switch VAE, copy current image, copy image to favorites
- Added new developer tool: Open CMD in SD Conda environment
- DreamBooth training: Images can now be automatically resized if they are not 512x512
- DreamBooth training: Added slider for steps multiplier (0.5x-2x)
- GUI improvements, prompt and negative prompt are now separate
- Image viewer now also shows "actual" image resolution (for upscaled images)
- Sliders now also allow you to type a value instead of dragging the handle
- Loading image metadata now also works for images generated with automatic1111
- "Delete All Current Images" now requires a confirmation
- Improved prompt history/queue UI (show full prompt on hover, and more)
- Improved GUI rendering, should have less flickering now
- Fixed bug where model pruning would say "Failed..." even if it was successful
SD GUI 1.6.0 Changelog:
- Added Dreambooth training (24 GB VRAM required for now!)
- Added support for prompt emphasis - () for more, {} for less
- Added Model Quick Switcher: Press Ctrl+M to quickly change the current AI model
- Added model folder manager: You can now add additional model folders to load models from
- Pop-Up Image Viewer: When slideshow mode is enabled, Left/Right arrow keys change images
- Pop-Up Image Viewer: Window can now be resized/zoomed in 25% steps using mouse wheel
- Pop-Up Image Viewer: Added "Always on Top" option to keep window always on top
- Added lots of hotkeys, documented on Github Readme
- Words in prompt field can now be deleted using CTRL+BACKSPACE, like in most text editors
- Model Pruning Tool: Added option to delete input file if pruning was successful
- Fixed an issue where the Stable Diffusion process would be killed when cancelling
- Fixed prompt queue not working after running first entry
SD GUI 1.5.0 Changelog:
- SD Update: Supports exclusion words, runs on 4 GB VRAM (when no other apps are open)
- UI is now more flexible, window can be resized, prompt field is bigger and has zoomable text
- Added CodeFormer face restoration as an alternative to GFPGAN
- Updated RealESRGAN (upscaler), should now be faster with same or better quality
- Added button to delete either the current image, or all generated images
- Added separate checkboxes for including prompt/seed/scale/sampler/model in filename
- Added option to save original image in addition to the post-processed image (if postproc is enabled)
- Added option to select the CUDA device (Automatic, CPU, or a specific GPU)
- Added model merging tool
- Added model pruning tool (strip EMA data and/or convert to fp16 half-precision for 2 GB models)
- Added option to unload Stable Diffusion after each generation (like in pre-1.4.0)
- Added reliable orphan process handling (Python no longer stays in RAM if the GUI crashes)
- Image Viewer slideshow: Added short cooldown after using prev/next image buttons
- Image Viewer context menu: Added button to re-generate single image with current settings/seed
- Image Viewer Pop-up: Now borderless, 100% zoom by default, double-click for fullscreen
- Image Viewer Pop-up: Added Slideshow Mode which mirrors the regular image viewer when enabled
- Added image load form, allows you to use as init image, load settings from metadata, or copy prompt
- Images can now be loaded from clipboard, not just from files
- Prompt History: Added option to disable history, added text filter
- You can now add an entry to the prompt queue by right-clicking on its icon
- Disabled post-processing with Low Memory Mode as it was not working properly
- Prompt text in filenames now strips weighting (won't create new folders for each weight change)
- Current model name gets printed whenever Stable Diffusion is started
- Full Precision is now enabled by default on GTX 16 series cards to fix compatibility with them
- Fixed empty/invalid prompts (e.g. newlines) counting towards the target image amount
- Some fixes regarding cancelling the generation process and handling crashes