What is this workflow?

This is my personal ComfyUI workflow built upon what I've learned and used in SD1.5, expanded to cover all possible use cases for myself. On a 4GB VRAM + 16GB RAM system, with everything active, it can still run and provide strong results (if you're willing to wait a while).

This workflow includes so, so many things that it's better to view it for yourself. The following is a small subset of supported features of this workflow:

• v-prediction support + RescaleCFG (disabled by default)

• Controllable CLIP Skip

• ELLA + ollama prompt upscale (SD1.5 exclusive feature)

• Scalable ControlNet group (disabled by default)

• Scalable IP-Adapter group (disabled by default)

• Dynamic Thresholding (disabled by default)

• 2-pass txt2img

• Perlin noise latent blend

• Watermark removal using CLIPSeg + Lama cleaner

• Scalable ADetailer group

• Notifications and sounds

• Preview chooser for batch images

• Full Civitai metadata support

• Prompt and LoRA scheduling

• Multi-checkpoint setup (still in testing)

• Wildcard (and wildcard file) support

• ...

Despite being built for SD1.5 first and foremost, this workflow can also use SDXL models or any all-in-one checkpoint models easily that is supported by prompt-reader-node. In these cases, ELLA would be completely useless, so do not enable it.

How to use this workflow?

• Step 0: Get ComfyUI and ComfyUI-Manager

• Step 1: Download the workflow file

• Step 2: Import it into ComfyUI

• Step 3: Download all missing nodes

• Step 4: Grab all missing models (more details below)

• Step 5: Have fun generating!

Model requirements

Basic

• A checkpoint file (SD1.5, SDXL,...)

• A VAE file (optional if baked VAE)

• 4x-AnimeSharp upscale model (or any other 4x upscale model of choice)

• 2x-AniScale2 upscale model (or any other combination of upscale model for the final upscale step)

ControlNet

All models are available inside ComfyUI-Manager under Model Manager. If you're using PonyXL or IllustriousXL, adapt your own solutions to it.

• Depth ControlNet (using Depth-Anything preprocessor)

• Lineart ControlNet (using AnyLine Lineart preprocessor)

• OpenPose ControlNet (using DWPose preprocessor)

• Depends on what other ControlNet model you add

Other features

• IP-Adapter: an IP-Adapter model and its related CLIP-G (CLIP Vision) model. They should all be available under Model Manager

• ELLA: See https://github.com/TencentQQGYLab/ComfyUI-ELLA?tab=readme-ov-file#orange_book-models

• ADetailer/FaceDetailer: Most models should be available under Model Manager, except Anzhc's face YOLO, which can be acquired at https://huggingface.co/Anzhc/Anzhcs_YOLOs. Place the downloaded file inside ultralytics/segm

• Ollama: See https://ollama.com/download and follow their instructions. After installing ollama on your system, grab https://ollama.com/huihui_ai/llama3.2-abliterate:3b