Abominable Workflows is a collection of ComfyUI workflows that unleashes the full potential of PixArt-Sigma by enhancing its exceptional prompt-following capabilities with the rich detail of SD15 models.
Each workflow generates a distinct style of imagery:

• abominable_PHOTO_ : Realistic images with photographic details.
• abominable_DARKFAN80_ : Dark, cinematic 80s style with VHS aesthetics and dramatic lighting.
• abominable_PIXEL_ : Pixel art style with retro aesthetics and vibrant, blocky shapes.
• abominable_INK_ : Ink-style illustrations with bold outlines and a hand-painted finish.
• abominable_1GIRL_ : Images centered on a captivating woman with photographic realism.
• abominable_MILO_ : European stylized comic book aesthetic with intricate linework.
• classic_abominable_spaghetti : Similar experience to the previous v2 but with all the new enhancements.

• Required Files
• Required Nodes
• How to Automatically Install the Required Nodes
• Manually Installing Required Nodes on Linux
• Manually Installing Required Nodes on Windows

To use the workflows, you need to have the following models installed in ComfyUI.
Place them in the corresponding directories as specified below.

<your_comfyui_dir> / models / checkpoints :

• pixart_sigma-FP16.safetensors (1.2GB)
• photon_refiner-FP16.safetensors (2.1GB)

<your_comfyui_dir> / models / clip :

• t5_xxl_encoder-FP8.safetensors (4.9GB)

<your_comfyui_dir> / models / vae :

• pixart_vae.safetensors (0.1GB)

Additionally, the workflows require the following custom nodes to be installed:

• ComfyUI_ExtraModels: provides support for PixArt-Sigma.
• ComfyUI-Crystools: used for some simple string operations.

The easiest way to install the required nodes is by using ComfyUI-Manager. This extension for ComfyUI offers management functions to install, remove, disable, and enable custom nodes. It simplifies the process and can save you a lot of trouble.

If for some reason you cannot use ComfyUI-Manager, follow the instructions below to install the nodes manually.

To manually install the nodes, open a terminal and run the following commands:

cd <your_comfyui_dir>
git clone https://github.com/city96/ComfyUI_ExtraModels ./custom_nodes/ComfyUI_ExtraModels
git clone https://github.com/crystian/ComfyUI-Crystools ./custom_nodes/ComfyUI-Crystools

If ComfyUI is using a virtual environment, you must activate it before installing the dependencies:

# You might need to replace '.venv' with the directory
# where the virtual environment is located
source .venv/bin/activate

Then, install the dependencies required by the nodes:

python -m pip install -r ./custom_nodes/ComfyUI_ExtraModels/requirements.txt
python -m pip install -r ./custom_nodes/ComfyUI-Crystools/requirements.txt

If you are using the standalone ComfyUI release on Windows, open a CMD in the "ComfyUI_windows_portable" folder (the one containing your run_nvidia_gpu.bat file).

From that directory, run the following commands to install the required nodes:

git clone https://github.com/city96/ComfyUI_ExtraModels ComfyUI/custom_nodes/ComfyUI_ExtraModels
git clone https://github.com/crystian/ComfyUI-Crystools ComfyUI/custom_nodes/ComfyUI-Crystools

Then, install the dependencies required by these nodes:

.\python_embeded\python.exe -s -m pip install -r .\ComfyUI\custom_nodes\ComfyUI_ExtraModels\requirements.txt
.\python_embeded\python.exe -s -m pip install -r .\ComfyUI\custom_nodes\ComfyUI-Crystools\requirements.txt

• Editable parameters grouped together.
• Option to select portrait or landscape image orientation.
• Option to choose between two samplers (fast/quality).
• Option for refiner to ignore prompt and focus on other elements.
• Variations by changing the refiner seed.
• Preconfigured workflows with different styles.
• Unified PixArt prompt and refiner prompt.
• CivitAI prompt extractor compatibility.
• Support for PixArt 2K model.
• Automatic detection of the model used and the optimal image size.

• Explanation of how to install required files.
• Explanation of how to manually install nodes on Linux and Windows.
• Models in FP16 or FP8 for minimal VRAM (6 or 8GB).
• Apply ComfyUI advances to speed up inference (own nodes?).
• Simplify installation with minimal custom nodes.

Copyright (c) 2024 Martin Rizzo
This project is licensed under the MIT license.
See the "LICENSE" file for details.

I would like to thank the developers of PixArt-Σ for their outstanding work. Their model has been a crucial component in the development of my workflows.

Further Information about PixArt-Σ:

• PixArt-Σ GitHub Repository
• PixArt-Σ Hugging Face Model
• PixArt-Σ arXiv Report