-
Notifications
You must be signed in to change notification settings - Fork 96
/
changelog-motd.txt
72 lines (63 loc) · 4.34 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
SD GUI 1.11.0 Changelog:
- 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
SD GUI 1.10.0 Changelog:
- New: Support for Stable Diffusion 2.x models
- New: Added outpainting (expanding images into any direction)
- New: Built-in Self-Updater (experimental)
- New: Added Karras samplers, including DPM++ 2M Karras
- New: Image generation symmetry options
- New: Utility to directly download Huggingface Diffusers models (no need for git clone etc.)
- New: Option to skip last CLIP layers ("CLIP Skip") which can improve quality with some models
- New: Chroma key option when loading an image (e.g. turn black pixels transparent)
- New additional config file "settings.ini" where power users can adjust certain values
- Improved: Loading transparent images from clipboard is now possible
- Improved: Any model format can now be loaded directly, Diffusers (fastest), Safetensors, Ckpt
- Improved: Prompts can now be longer than 77 tokens (experimental)
- Improved: Textual Inversion embeddings are now properly integrated, newer formats supported
- Improved: InstructPix2Pix and SD ONNX no longer need to reload the model on each run
- Improved: Maximum resolution and prompt history limit have been increased
- Improved: UI is now split into categories that can be collapsed
- Improved: A wide range of UI/UX performance improvements, reduced flickering, etc
- Too many bug fixes to list
SD GUI 1.9.1 Changelog:
- 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
Notes:
- InstructPix2Pix supports resolutions divisible by 8, not only 64 like regular SD
- SD 2.x models are not yet supported, scheduled for next major update
SD GUI 1.9.0 Changelog:
- New: InstructPix2Pix (Settings -> Image Generation Implementation -> InstructPix2Pix)
- New: Option to show the input image next to the output for comparisons
- New: 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