Cagliostro Colab UI

All-in-One, Customizable and Flexible AUTOMATIC1111's Stable Diffusion Web UI for Google Colab.

What's New?

v.3.0.0 (10/06/23)

• Rewrote cagliostro-colab-ui codebase from scratch.

  • Used in-house module Colablib as the primary library for go-to functions, such as colored print, git function, download syntax, etc.
  • Added numerous trivial but important pieces of information, such as the Python version, torch version, current commit hash.
  • Improved console logs, using print_line() and cprint() from Colablib.

• Merged output_to_drive with mount_drive.

  • Now, the output path is automatically set to drive if Google Drive is mounted.

• Built-in wildcard support.

• Renamed dpm_v2_patch to dpmpp_2m_v2_patch.

• Added new section: Optimization config.

  • Introduced mobile_optimizations to keep Colab tab alive for mobile users.
  • Removed Keep Tab Alive for Mobile due to similar functionality with the new optimization.

• New extensions!

  • ilian6806/stable-diffusion-webui-state : Preserves Web UI parameters (inputs, sliders, checkboxes, etc.) after page reload.
  • pkuliyi2015/multidiffusion-upscaler-for-automatic1111 : Offers tiled Diffusion and VAE optimization.
  • Zuellni/Stable-Diffusion-WebUI-Image-Filters : A simple image postprocessing extension using the Pillow library.
  • Linaqruf/Umi-AI-debloat : Wildcard manager, a fork of Tsukreya/Umi-AI-debloat that has already fixed the lowercase problem.
  • AlUlkesh/sd_delete_button : Adds a delete button for Automatic1111 txt2img and img2img.

• New default models!

  • Added AnyLoRA_Anime_Mix as a new AnyLoRA variant for 2D Anime generation, similar to NovelAI.
  • Renamed the old AnyLoRA to AnyLoRA_Default.
  • Added Ghost_Note_Delta, SDHK_V3, for another Anime Model.
  • Added Majic_Mix_V5 to replace Chillout Mix.
  • Replaced replicant_v2 with the latest and improved version, Replicant_V3.
  • Reintroduced Illuminati Diffusion V1.1.
  • Removed Waifu Diffusion 1.5 models.

• New default VAEs!

  • Added Blessed VAE.
  • All default VAEs are now in .safetensors! Thanks to NoCrypt.

• ControlNet

  • Added new Annotator, up to Lama cleaner.
  • Included a list of new SDv2.x ControlNet Models from thibaud/controlnet-sd21.
  • Added Custom ControlNet Model section to download custom controlnet models such as Illumination, Brightness, the upcoming QR Code model, and any other unofficial ControlNet Model.
  • Please ensure your custom ControlNet model has sd15/sd21 in the filename.
  • Reintroduced t2i_adapter_model.

• Custom Download Corner

  • Added instructions for using the fuse: prefix, handling multiple URLs, and loading models from Google Drive.

  • The prune:, fp16:, or fp32: prefixes currently are not available.

  • Introduced a new section: Download From Textfile.

    • This uses a similar approach to the etherealxx/batchlinks-webui extensions.
    • Provide a custom download URL for a .txt file instead of using the URL field. Edit the file: /content/download_list.txt.
    • Available hashtags: #model, #vae, #embedding, #lora, #hypernetwork, #extensions, #upscaler. Aliases are not currently supported.
    • Alternatively, you can input your .txt file into custom_download_list_url. This works for pastebin.

    Example:

    #model
    url1
    url2
    
    #lora
    url1 | filename1
    
    #embedding
    fuse:path

• Launch

  • Added --opt-sdp-mem-attention option for accelerator.
  • Included error handling if the user forgets to download the model and vae. It will automatically download these two files before launch:
    • Model: AnyLoRA_Anime_Mix
    • VAE: Animevae
  • Added Token Merging Ratio or ToME SD and Negative Guidance Scale to quicksettings

• Refactored Download Generated Images and other Extras cells by wrapping code inside functions.

• Bugfixes:

  • Fixed dpmpp_2m_v2_patch, now it's working properly.
  • Temporary fixed LoRA not applied when using Hires Fix
  • Fixed file extensions undefined in Custom Download Corner

v.2.6.2 (17/05/23)

• Bugfixes:
  • Attempting to fix generate button stuck problem by adding latest_gradio option to force update gradio to 3.31

v.2.6.1 (17/05/23)

• Added sd-civitai-browser extension back, but this time using SignalFlagZ/sd-civitai-browser version.
• Changed stable-diffusion-webui-latent-two-shot extension with ashen-sensored's fork.
• Updated alternative tunnels recommendation: ngrok > gradio > cloudflared > remotemoe > localhostrun > googleusercontent. The ranking is based on the following reasons:
  • Gradio: Faster and slightly more stable than the others, but may have queue problems.
  • Cloudflared: Slow to start the application, loads extra networks and images.
  • Remotemoe: Currently not usable.
  • Localhostrun: Often gets disconnected, requiring a forced restart of cells to make it work.
  • Googleusercontent: -
• Bugfixes:
  • Fixed stable-diffusion-webui-composable-lora not being installed in repo_type values: ["AUTOMATIC1111", "AUTOMATIC1111-dev"].
  • Set commit_hash to empty and use latest commit because the bugs already fixed.

v.2.6.0 (16/05/23)

• Updated to the latest commit.
• Available repo_type values: ["AUTOMATIC1111", "AUTOMATIC1111-dev", "Anapnoe"].
• Updated the new tile model with control_v11f1e_sd15_tile_fp16.safetensors.
• Added 2 optimization options:
  • ram_alloc_patch, also known as the Camenduru Patch, to decrease RAM usage.
  • colab_optimizations, also known as TheLastBen Patch, to load the model in VRAM and fix the gradio queue problem.
• Added the use_presets option to turn on or off the default prompt.
• Simplified the Arguments category options.
• Set a default value for gradio_auth.
• Changed yfszzx/stable-diffusion-webui-images-browser to zanllp/sd-webui-infinite-image-browsing.
• Added the extension butaixianran/Stable-Diffusion-Webui-Civitai-Helper.
• Added the extension Bing-su/adetailer.
• Added the extension canvas-zoom.
• Added negative embeddings EasyNegativeV2.
• Added a new theme for Anapnoe-webui: ["minimal", "minimal-orange"].
• Bug Fixes:
  • Fixed Inpaint not showing images #28. It looks like Bing-su/adetailer and canvas-zoom are unusable in Anapnoe-webui.
  • Fixed the quicksettings problem after the master repo was migrated from strings to a list. #27 #29.
  • An annoying bug has recently occurred in Anapnoe-webui. Extra networks are not displaying Models, Embeddings, LoRA, etc., even if they exist in the respective folder. Temporary fixed by reset commit hash to f2b9c2c. Track discussion here

v.2.5.3 (21/04/23)

• AUTOMATIC1111's stable-diffusion-webui has been removed because Google Colab has prohibited the usage of any string named stable-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 the Anapnoe UI integrated with gradio 3.23.0.
• sd-webui-tunnels forked by camenduru 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.

v.2.5.2 (19/04/23)

• Reformat and simplified most cells
• Rewording all available variable to make it easier to read, example: git_pull to update_webui
• Added output_drive_folder to customize gdrive outputs folder name
• Removed clean_install option
• Merged load_v2_in_vram and merge_in_vram to colab_optimization and set to False
• Deleted stable_diffusion_v_1_5 and replicant_v1 from available models
• Added custom_upscaler_url back

v.2.5.1 (18/04/23)

• Added cldm_config.yaml for every ControlNet model (refer to link)
• Removed custom_control_url from Custom Download Corner cell
• Removed illuminati_diffusion_v1_1 from Available 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 the Arguments list.

v.2.5.0 (16/04/23)

• 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 to True, skip updating stable-diffusion-webui-images-browser when update_extensions set to True
• 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 to sd21_control_v10_sd21_model
  • Renamed wd15_control_model to wd15_control_v10_sd21_model
• Revamped how Custom Download Corner works.
  • Removed custom_upscaler_url, custom_control_url, and custom_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.

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 to cagliostro, and the password is a randomly generated 6-character combination of ASCII letters and numbers.
• Added an option to enable --opt-sdp-attention instead of xformers to accelerate PyTorch 2.0.

v.2.2.0 (03/04/23)

• Update Web UI to the latest version
• Update xformers to 0.0.18
• git checkout for extension using Gradio 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

v.2.1.0 (31/03/23)

• 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 of git remote origin set-url and git reset --hard
• Added Theme Selector section to changed default (anapnoe-webui) theme before starting Web UI

v.2.0.1 (27/03/23)

• Clean install and update Web UI to latest version, commit hash : 955df7751eef11bb7697e2d77f6b8a6226b21e13
• Manually edit config.json and ui-config.json
• Change auto_model and auto_vae logic, change config.json instead of using --vae_path and --ckpt

v.2.0.0 (27/03/23)

• Reformat notebook with black python formatter
• Update Web UI and all extensions version (not latest)
• Downgrade xformers to 0.0.16
• Downgrade triton to 2.0.0
• Set both lora_dir and additional_networks_extra_lora_path path to os.path.join(repo_dir, "models/Lora")
• Set lora_dir path to os.path.join(repo_dir, "models/ControlNet")
• Reset Anapnoe UI commit hash to 802fb8a16ba4bdbfba0dca55f9cdb265f4bd86f2
• Force reset commit hash to 3b47b000199ea8baf724080936ef985f53b3d081 when use_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
  • Anything V3.0
  • AnyLoRA
  • Anime Pastel Dream
• Deleted 3 default model
• Anything V3.2
• Anything V3.3
• HoloKuki V2
• Changed Chillout-Mix to Chillout-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 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

v.1.4.1 (11/03/23)

• 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

v.1.4 (09/03/23)

• 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 and sd-webui-llul

v.1.3 (01/03/23)

• 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 to 3cd625854f9dc71235a432703ba82abfc5d1a3fc when try_new_ui_ux set to True, as it's a stable commit history for now.
• Added t2iadapter_sketch-fp16.safetensors to t2iadapter model list.

v.1.2.2 (23/02/23)

• 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.

v.1.2.1 (21/02/23)

• 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.