Caution

This README was generated by AI because the author is too lazy to write documentation. Deal with it. 🤷

ComfyUI Workflows Collection

A collection of optimized ComfyUI workflows for high-quality AI image generation using the Illustrious model family.

---

📦 Workflows

1. txt2img_simple_gen.json - Simple Text-to-Image Generator

A streamlined workflow for quick, high-quality image generation.

Features:

• Single-pass KSampler generation
• LoRA support with Power Lora Loader
• Optional 4x upscaling with UltraSharpV2
• Configurable aspect ratios (SDXL presets)
• Automatic save with date-based organization

Default Configuration:

Parameter	Value
Resolution	1216×832 (3:2 landscape)
Steps	25
CFG Scale	4
Sampler	euler_ancestral
Scheduler	normal
CLIP Layer	-2

Node Flow:

Checkpoint → LoRAs → CLIP Encode → KSampler → VAE Decode → Save/Upscale

---

2. txt2img_full_detailer.json - Full Detailer Pipeline

An advanced workflow with multi-stage face/body refinement and hi-res upscaling.

Features:

• Dual FaceDetailer passes for enhanced facial quality
  • First pass: Face detection with 99coins_anime_girl_face_m_seg.pt
  • Second pass: NSFW body refinement with ntd11_anime_nsfw_segm_v5.pt
• Differential Diffusion for smooth inpainting transitions
• Hi-Res upscaling pipeline:
  • 4x UltraSharpV2 model upscale
  • Secondary KSampler refinement pass
• Smart image routing via ImpactSwitch
• Multiple save points for workflow inspection

Default Configuration:

Parameter	Value
Base Resolution	1216×832
Final Hi-Res	2432×1664
Base Steps	30
Hi-Res Denoise	0.15
Face Detailer Denoise	0.7 / 0.3

Node Flow:

┌─────────────────────────────────────────────────────────────────────────┐
│ Checkpoint → LoRAs → KSampler → VAE Decode → Save Original              │
│                                      │                                  │
│                                      ▼                                  │
│                              FaceDetailer #1 → Preview                  │
│                                      │                                  │
│                                      ▼                                  │
│                              FaceDetailer #2 → Preview                  │
│                                      │                                  │
│                                      ▼                                  │
│                      ┌──────── ImpactSwitch ────────┐                   │
│                      ▼                              ▼                   │
│              ImageUpscale (4x)              (Direct path)               │
│                      │                                                  │
│                      ▼                                                  │
│              ImageScale → VAEEncode → KSampler → VAEDecode → Save Hi-Res│
└─────────────────────────────────────────────────────────────────────────┘

---

🛠️ Required Custom Nodes

Node Pack	Purpose
ComfyUI-Impact-Pack	FaceDetailer, ImpactSwitch, DifferentialDiffusion
ComfyUI-Impact-Subpack	UltralyticsDetectorProvider
rgthree-comfy	Power Lora Loader
ComfyUI_JPS-Nodes	Text Prompt nodes
ComfyUI_Comfyroll_CustomNodes	SDXL Aspect Ratio

---

📋 Required Models

Checkpoint

• illustrij_v18.safetensors - Illustrious anime model

LoRAs

• Dramatic Lighting Slider.safetensors - Enhances lighting effects
• Hyperrealistic_illustrious.safetensors - Adds realistic details

Upscale Model

• 4x-UltraSharpV2.safetensors - High-quality 4x upscaler

Detection Models (for Full Detailer)

• segm/99coins_anime_girl_face_m_seg.pt - Anime face segmentation
• segm/ntd11_anime_nsfw_segm_v5.pt - Body segmentation

---

📁 Output Structure

Images are saved with date-based organization:

ComfyUI/output/comfyui/
└── dd-MM-yyyy/
    ├── original_*.png       # Initial generation
    └── finalhires_*.png     # After detailing + upscale

---

🚀 Quick Start

1. Install all required custom nodes via ComfyUI Manager
2. Download and place the required models in their respective directories
3. Import the desired workflow JSON into ComfyUI
4. Enter your prompt in the positive text box
5. (Optional) Add negative prompts for unwanted elements
6. Click Queue Prompt to generate

---

💡 Tips

• Simple workflow is best for rapid iteration and testing prompts
• Full detailer is ideal for final renders requiring maximum quality
• Both workflows use LoRA strength 0.7 by default - adjust as needed
• The hi-res pass uses low denoise (0.15) to preserve details while adding sharpness

---

📝 License

These workflows are provided as-is for personal and educational use.