All-in-One, Customizable and Flexible AUTOMATIC1111's Stable Diffusion Web UI for Google Colab. Inspired by Linaqruf project
AUTOMATIC1111's stable-diffusion-webui
has been removed because Google Colab has prohibited the usage of any string namedstable-diffusion-webui
due to its massive usage of webui.Anapnoe UI
has been set as default forcibly, also added warning messages.- An
experimental
section has been added to install theAnapnoe UI
integrated withgradio 3.23.0
. sd-webui-tunnels
forked bycamenduru
has been removed since the new ToS of Colab prohibits SSH shell or any similar tools.- Now, there are two versions of the notebook available:
-
Cagliostro Colab UI: A compact and lightweight version that removes
AUTOMATIC1111's stable-diffusion-webui
. -
Cagliostro Colab UI Pro: It has all the features of the Cagliostro Colab UI update in 19/04. However, since I do not have a Colab Pro subscription, I am unlikely to maintain it.
-
- Reformat and simplified most cells
- Rewording all available variable to make it easier to read, example:
git_pull
toupdate_webui
- Added
output_drive_folder
to customize gdrive outputs folder name - Removed
clean_install
option - Merged
load_v2_in_vram
andmerge_in_vram
tocolab_optimization
and set toFalse
- Deleted
stable_diffusion_v_1_5
andreplicant_v1
from available models - Added
custom_upscaler_url
back
- Added
cldm_config.yaml
for every ControlNet model (refer to link) - Removed
custom_control_url
fromCustom Download Corner
cell - Removed
illuminati_diffusion_v1_1
fromAvailable SDv2.x Model
list as the model has become exclusive to a certain image generation website - Reworded certain arguments to make them more readable
- Removed
medvram
option from theArguments
list.
- Update Web UI to the latest version
- Moved
What's new?
to GitHub repository - Added link to Cagliostro Colab UI User Manual
- Added
cheatsheet
in every header and subheader based on Cagliostro Colab UI User Manual explanation - Reformatted Notebook to be more readable.
- If
use_anapnoe_ui
set toTrue
, skip updatingstable-diffusion-webui-images-browser
whenupdate_extensions
set toTrue
- Added 2 more
os.environ
changes- os.environ["PYTHONDONTWRITEBYTECODE"]='1'
- os.environ['PYTHONWARNINGS'] = 'ignore'
- Added Replicant V2.0 as new SDv2.x model
- ControlNet V1.1 Update!
- Removed old Annotator and ControlNet V1.0 model, and added new ones. Total: 13 new Annotator and 14 new ControlNet V1.1 models.
- Renamed
sd21_control_model
tosd21_control_v10_sd21_model
- Renamed
wd15_control_model
towd15_control_v10_sd21_model
- Revamped how
Custom Download Corner
works.- Removed
custom_upscaler_url
,custom_control_url
, andcustom_components_url
- Added a feature to prune model, fuse folder, copy from Google Drive. Users can find how to use it in this link or this table.
- Removed
Feature | Description | How to Use | Example |
---|---|---|---|
Multiple Downloads | Download multiple files at once. | Fill in the URL fields with the links to the files you want to download. Separate multiple URLs with a comma. | url1, url2, url3 |
Auto-prune | Prune models after downloading | Add fp16: or fp32: before URLs. |
fp16:url1 |
Copy from Google Drive | Copy models from Google Drive and load them in the session. | Make sure you have already mounted Google Drive. Type the path to your model/lora/embedding from Google Drive. | /content/drive/MyDrive/path/to/folder |
Fusing Folder | Fuse models/embeddings/LoRA folder to /content/fused/{category} . |
Make sure you have already mounted Google Drive. Add fuse: before the path to the folder. |
fuse:/path/to/gdrive/folder |
Auto-extract | Extract files after downloading. | Add links/to/file ending with .zip or .tar.lz4 . Extract files to specified destination directory. |
https//link.com/to/file |
Install Extensions | Install extensions for Stable Diffusion Web UI. | Add the link to the GitHub repository to custom_extension_url . |
https://github.com/user/repo |
- Changed the Gradio authentication logic to use a
boolean
instead. The username is set tocagliostro
, and the password is a randomly generated 6-character combination of ASCII letters and numbers. - Added an option to enable
--opt-sdp-attention
instead ofxformers
to accelerate PyTorch 2.0.
- Update Web UI to the latest version
- Update xformers to
0.0.18
git checkout
for extension usingGradio 3.23
or above- If using Anapnoe UI/UX, users can't
update_extensions
at the moment to prevent code conflict because of different gradio version
- Added Reinstall logic
- Fix bugs when upload images to ControlNet canvas, the images stretched
- Fix bugs when Web UI can't starting if
use_anapnoe_ui
set to True - Changed
use_anapnoe_ui
logic by downloading and unpacking separate repo instead ofgit remote origin set-url
andgit reset --hard
- Added Theme Selector section to changed default (anapnoe-webui) theme before starting Web UI
- Clean install and update Web UI to latest version, commit hash :
955df7751eef11bb7697e2d77f6b8a6226b21e13
- Manually edit
config.json
andui-config.json
- Change
auto_model
andauto_vae
logic, changeconfig.json
instead of using--vae_path
and--ckpt
- Reformat notebook with black python formatter
- Update Web UI and all extensions version (not latest)
- Downgrade
xformers
to0.0.16
- Downgrade
triton
to2.0.0
- Set both
lora_dir
andadditional_networks_extra_lora_path
path toos.path.join(repo_dir, "models/Lora")
- Set
lora_dir
path toos.path.join(repo_dir, "models/ControlNet")
- Reset Anapnoe UI commit hash to
802fb8a16ba4bdbfba0dca55f9cdb265f4bd86f2
- Force reset commit hash to
3b47b000199ea8baf724080936ef985f53b3d081
whenuse_anapnoe_ui
is enabled - Added
mount_drive
function - Added more error handling
- Set some environment variables:
- os.environ['colab_url'] = eval_js("google.colab.kernel.proxyPort(7860, {'cache': false})")
- os.environ["LD_PRELOAD"] = "libtcmalloc.so"
- os.environ["TF_CPP_MIN_LOG_LEVEL"] = "3"
- os.environ["SAFETENSORS_FAST_GPU"]= '1'
- Added 3 new default model
- Deleted 3 default model
- Anything V3.2
- Anything V3.3
- HoloKuki V2
- Changed
Chillout-Mix
toChillout-Mix-Ni
- Changed
Illuminati V1.1
download link from CivitAI to Huggingface - Put all ControlNet things to new separated cell
- Added
pre_download_annotator
back - Added
sd21_control_model
- Added
wd15_control_model
- Added option to change ControlNet model and Adapter config outside Web UI
- Added
- Added support for load custom url from Google Drive, tricky, by copying from Google Drive to destination path.
- Set
--multiple
as default alternative tunnels, please don't use Gradio URL at the moment- Recommended Tunnels:
ngrok
>cloudflared
>remotemoe
>localhostrun
>googleusercontent
>gradio
- Trying to establish my own style, deleted all possible NoCrypt and TheLastBen's code legacy
- Temporary removed
pre_download_annotator
- Deleted
PastelMix
from available models because the model license sold to fantasy.ai - Deleted
RefSlave-V2
from available models because the author removed his model from huggingface and civitai - Added HoloKuki V2 as to available models
- Update t2i adapter model, currently has 8 model
- Update xformers pre-compiled wheels to
xFormers 0.0.17.dev466
- Update pre-installed dependencies for
Python 3.9.16
- Added new Stable Diffusion Web UI Extensions:
batchlinks-webui
andsd-webui-llul
- Added an option to save outputs to drive.
- Moved support button to separated and hidden section, because it looks ugly.
- Fixed some bugs where VAE can't be changed in webui.
- Updated dependencies, webui, and extensions to the latest version.
- Deleted
hitokomoru_v1_5
. - Added WD 1.5 Beta 2 - Aesthetic. Releases note: here
- Added
illuminati_diffusion_v1_1
. Required embeddings coming soon. - Added
ref_slave_v2
and set it to default - Used
git reset
to move the head to3cd625854f9dc71235a432703ba82abfc5d1a3fc
whentry_new_ui_ux
set to True, as it's a stable commit history for now. - Added
t2iadapter_sketch-fp16.safetensors
to t2iadapter model list.
- Added Extra section for optional cell, the first cell added to the section is Download Generated Images V2, to store your output to huggingface and download it.
- Changed how the installation works. If the folder exists, then skip unpacking.
- Added
Replicant V1.0
as default model. - Added new UI/UX theme from Anapnoe.
[Experimental]
- Added support for
multi-controlnet
by including a slider in certain cells (default value:2
). - Backed up all ControlNet annotator and model data to a personal repository.
- Added
tqdm
to track installation, unpacking, and download progress since original logs are disabled. - Pulled the latest version of the repository and built-in extensions.
- Pre-downloaded ControlNet annotator/preprocessor.
- Added T2I Adapter model from TencentArc.
- Added several good custom upscalers such as
4x-Animesharp
,4x-UltraSharp
,Lollypop
, and others. - Added Video Loopback extension for creating videos in Img2img, now with ControlNet support.
- Deleted CivitAI browser extension because another extension called
sd-filer
serves the same purpose. You can download models, lora, and embeddings from the web UI. - Added Katanuki extension to convert results to transparent images.
- Deleted Haku Img Extensions because they were rarely used.
- Added 'Use Old Karras Scheduler' option to quick settings (header).
- Fixed an invalid URL for Chillout Mix Pruned and SD 1.5 Pruned.
- Fixed the 'Download Generated Images' cell when creating a duplicate zip from 'output.zip(1)' to 'output(n+1).zip'.
- Added checkboxes for randomly selecting VAE and Model.