# 🔱 Trinity SDXL Workflow (Illustrious / Pony / NoobAI / Img2Img / inpainting)
A powerful, modular, and highly optimized SDXL workflow designed for exceptional image quality and flexibility.
> Version 1.0
> This is the first public release of Trinity SDXL Workflow. Future updates may include new features, optimized stages, and improved settings.
Tested on:
- GPU: RTX 3070 Ti
- RAM: 32 GB
- CPU: Intel Core i5-9400F
- OS: Windows 11
This workflow is fully compatible with Illustrious, Pony, and NoobAI based models.
It features a multi-stage generation system, advanced detailing tools, and full support for Empty Latent, Img2Img, and Inpainting.
➖➖➖➖➖➖
### ✨ Key Features
- Multi-stage generation (1st → 2nd → 3rd Generation)
- Optional Person Detailer, Face Detailer, and Hand Detailer
- High-quality Upscaler + Skin Contrast enhancement
- Full support for New Image, Img2Img, and Inpainting
- Independent seed control for every generation stage
- Fast Group Bypasser for quick speed/quality switching
- Civitai Metadata Saver (fully compatible with Civitai)
- Highly modular structure — easily enable or bypass any stage
➖➖➖➖➖➖
### 📦 Required Custom Nodes
Install these custom node packs before using the workflow:
- Comfyroll Custom Nodes
- rgthree-comfy
- ComfyUI-Essentials
- comfyui_controlnet_aux
- Crystools
- ComfyUI-Image-Saver
- ComfyUI-Impact-Pack
- ComfyUI-Impact-Subpack
- ComfyUI-KJNodes
- ComfyUI_UltimateSDUpscale
- comfyui-gps-supplements
➖➖➖➖➖➖
### 📘 Workflow Stage Guide
You can bypass optional sections if needed.
Avoid bypassing the core generation stages unless necessary, as it may reduce final quality.
1. Loaders
Loads the base model, CLIP, VAE, LoRAs, embeddings, and all primary resources.
2. Prompt Processing
Processes positive and negative prompts and prepares conditioning data.
3. Tools (Resolution & Randomize)
Contains resolution settings, Fast Group Bypasser, and independent seed controls for 1st, 2nd, and 3rd Generation.
4. ControlNet (Optional)
Guides character pose and anatomy to reduce artifacts and improve structural consistency.
5. 1st Generation
Creates the initial image with overall composition, character, and environment.
6. 2nd Generation
Refines facial features, eyes, hands, skin texture, and expression using a secondary prompt.
7. 3rd Generation
Reinforces the original prompt and restores character personality while preserving previous improvements.
8. Person Detailer (Optional)
Detects people and refines the full body and anatomy (stronger than Face Detailer).
9. Face Detailer (Optional)
Detects and enhances every visible face in the image.
10. Hand Detailer (Optional)
Detects hands and corrects finger/hand anatomy while preserving the original pose.
11. Upscaler (Optional)
Increases resolution and improves sharpness, skin texture, and overall clarity.
12. Add Skin Contrast (Optional)
Applies subtle contrast enhancement for better skin definition.
13. Civitai Metadata Saver (Optional)
Saves the final image with complete embedded metadata (model, LoRAs, prompts, workflow, and settings) for full Civitai compatibility.
➖➖➖➖➖➖
### 📐 Recommended Resolutions
- 🖥️🖥️ Dual Monitor: 1704x480
- 🖥️ Desktop: 1280x720
- ❎ Square: 1280x1280 or 720x720
- 📱 Phone (19.5:9): 544x1176
Note: The final output resolution will be more than double the selected Empty Latent size.
➖➖➖➖➖➖
### ⚡ Speed vs Quality Guide
For significantly faster generation, apply the following changes:
1. ControlNet → Bypass
2. 2nd Gen Tools → Set ALL switches to Input 2
3. CR Image Input Switch → Set to Input 2
4. 3rd Generation → Bypass
5. Person Detailer → Bypass
6. Face Detailer → Bypass
7. Hand Detailer → Bypass
8. Upscaler → Bypass
9. Add Skin Contrast → Bypass
📌 Steps 2, 3, and 4 are linked together and should always be changed as a group.
Recommended balanced configurations:
- 2, 3, 4, 8
- or 2, 3, 4, 5, 8
Keeping the Civitai Metadata Saver enabled is recommended if you plan to upload images to Civitai.
➖➖➖➖➖➖
### 🔀 Latent Mode Guide (Img2Img & Inpainting)
Black Switch
- 1 = Empty Latent (Create new image)
- 2 = Img2Img
Purple Switch
- 1 = Empty Latent / Img2Img
- 2 = Inpainting
Typical Usage:
- New Image → Black = 1 | Purple = 1
- Img2Img → Black = 2 | Purple = 1
- Inpainting → Purple = 2 (only when using a mask)
➖➖➖➖➖➖
### ⚙️ Recommended Generation Settings
- Steps: 20–30
- CFG Scale: 3.0–5.0
- Sampler + Scheduler: Euler + Simple or DPM++ 2M SDE + Karras
- Denoise Strength:
- 1st Generation → 1.0
- Other stages → 0.18 – 0.40
- Seed: Fixed for testing | Randomize for new variations
Higher denoise values allow stronger changes but may alter composition.
Avoid changing all denoise values at the same time.
➖➖➖➖➖➖
### 📦 Models & LoRAs
I spent a lot of time testing different models and LoRAs. Below are the ones that delivered the best results:
Base Model
- AniMayhem - Pale Rider (Recommended)
https://civarchive.com/models/1984400/animayhem-pale-rider
📂 ComfyUI/models/checkpoints/
ControlNet
- Illustrious-XL ControlNet OpenPose
📂 ComfyUI/models/controlnet/
Upscalers
- 1xSkinContrast-High-SuperUltraCompact
📂 ComfyUI/models/upscale_models/
VAE (Selectable)
📂 ComfyUI/models/vae/
LoRAs (Recommended)
- Ri-mix - Style LORA [Illustrious + Anima]
- Detailer | Tool / Concept LoRA IllustriousXL
📂 ComfyUI/models/loras/
➖➖➖➖➖➖
### 📌 Important Notes
- Only the image saved by the Civitai Metadata Saver contains full metadata. Preview images do not.
- For best quality, enable the same quality and face-enhancement LoRAs in Stage 1 (Loaders) that you use in the 2nd Generation stage (with the same weights).
- This workflow is optimized for the default checkpoint. If you use a different model, you may need to adjust Steps, CFG, Sampler, and Denoise values.
➖➖➖➖➖➖
Enjoy creating with Trinity SDXL 🔱
✨ © Please credit "🔱 Trinity SDXL Workflow" when using or sharing this workflow.
Your support helps keep the Trinity brand growing and evolving. 💫
Description
v1.0
This workflow is primarily optimized for Illustrious-based models.
However, it also works well with other SDXL-based models such as NoobAI, Pony, and similar checkpoints.
It is designed to be flexible and support a wide range of styles, not limited to a single model.