forked from n00mkrad/text2image-gui
/
changelog-motd.txt
95 lines (83 loc) · 5.89 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
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