Z-Image-Turbo-AIO-Workflow

SeeSeeLP

1.3K

工具

[Img2Img] Detection [Img2Img] ControlNet Advanced [Txt2Img]

详情

下载文件 (1)

关于此版本

Z-Image-Turbo Img2Img Detection - By sphiratrioth666

Professional workflow with SAM2, SEGS detailing, model upscaler, and grain

🎯 Key Features:

✅ SAM2 Segmentation - Precise area detection
✅ SEGS Detailer - Selective detail enhancement
✅ Model Upscaler - High-quality upscaling
✅ Grain Addition - Natural photographic look
✅ Smart Resize - Downscale to 2K/4K for reasonable file size
✅ Multi-LoRA Support - 4 LoRA slots
✅ PURGE VRAM - Extensive cleanup
✅ Play/Stop System - Test before saving

📦 Required Downloads:

Models:

Z-Image-Turbo-AIO FP8/BF16
Z-Image-Turbo-Fun-Controlnet-Union.safetensors
Model Upscaler (any 4x upscaler like ESRGAN)
SAM2 Model - For segmentation
⚠️ ControlNet in: ComfyUI/models/model_patches/

Custom Nodes:

MXToolkit - Sliders
rgthree-comfy - LoRA stack
comfyui_controlnet_aux
LayerUtility - PURGE + effects
Impact Pack - SEGS detailer
SAM2 - Segmentation
comfyui_image_metadata_extension

⚠️ ComfyUI 3.77+ required!

⚙️ Settings:

Steps: 9 (adjustable)
CFG: 1.0 (adjustable)
Sampler: res_multistep or euler_ancestral
Scheduler: simple or beta
Guide Size: Adjustable (SAM2)
Max Size: Adjustable (SEGS)

🎨 Workflow Pipeline:

Input Image → ControlNet guidance
SAM2 Segmentation → Auto-detect regions
SEGS Detailer → Enhance face, hands, details
Model Upscale → 4x upscaling (10K resolution)
Add Grain → Natural photographic texture
Resize to 2K/4K → Reasonable file size (~10-20MB)

🔍 How to Use:

Upload input image
Choose preprocessor - ControlNet type
Load LoRAs - Up to 4 slots
Write prompt - Detailed description
Adjust sliders - CFG, Steps, Guide/Max size
Click PLAY - Full detailing pipeline runs
Review result - Check detailing quality
Click SAVE - Final 2K/4K output

🎯 What Makes It Special:

SAM2 Segmentation:

Automatically detects faces, hands, and important areas for selective enhancement.

SEGS Detailer:

Applies focused detail enhancement only where needed. Fixes common issues like:

Blurry faces
Unclear hands
Lost details in upscaling

Model Upscaler:

Uses high-quality model upscaler (ESRGAN-style) to reach 10K resolution.

Grain Addition:

Adds subtle photographic grain at 10K resolution for natural look. Blends enhanced areas seamlessly.

Smart Resize:

Downscales back to 2K (2048px) or 4K (4096px) for:

Reasonable file sizes (10-20MB vs 100MB+)
Easier sharing and storage
Still exceptional quality

🎛️ Slider Controls:

Guide Size - SAM2 detection sensitivity
Max Size - Maximum detail patch size
CFG - Guidance strength
Steps - Generation steps
Denoise - Detail enhancement strength
LoRA 1-4 - Individual LoRA strengths

💡 Advanced Tips:

For Best Results:

Use high-quality input (1024px+)
Let SAM2 auto-detect regions
Adjust Guide Size if detection is off
Higher Denoise = more detail changes
Grain at 10K resolution looks most natural

When to Use:

Final generations for portfolio
Professional work requiring max quality
Images with faces/hands needing detail
When you want photorealistic texture
Publication-quality outputs

Performance:

Takes longer than simple workflows (~20-60 seconds)
Uses more VRAM (PURGE helps!)
Worth it for final results

🔧 Downscale/BYPASS:

Workflow includes option to:

Downscale - Resize to 2K/4K (recommended)
BYPASS - Keep full 10K resolution (huge files!)

Most users prefer downscaled version for balance of quality and file size.

🧹 PURGE VRAM:

Multiple PURGE nodes throughout pipeline prevent VRAM buildup during complex processing.

Credits: Advanced detailer workflow by sphiratrioth666
Based on: Z-Image-Turbo ControlNet workflow by SeeSeeLP
Version: Advanced NQ + ControlNet + SEGS Detailer Edition
Updated: December 2024

For professionals who need maximum quality and detail! ✨

"A bit over the top, but it works very well" - sphiratrioth666

模型描述

🚀 Z-Image-Turbo Advanced Workflows

By sphiratrioth666

Enhanced workflows with sliders, multi-LoRA, VRAM management, and professional detailing.

work - these are serious upgrades! 🚀

🎉 Advanced Workflows by sphiratrioth666

These workflows were modified and enhanced by user sphiratrioth666, based on the original Z-Image-Turbo-AIO workflows.

He integrated meaningful improvements and created an entirely new DETAILER workflow with SAM2 + SEGS!

Big thanks and full credits go to him! 🙏

Check out his work - these are serious upgrades! 🚀

📦 Three Workflow Variants:

1. 🎨 Txt2Img (Advanced)

Pure text-to-image with advanced controls

Key Features:

Interactive sliders (CFG, Steps, Denoise, Upscale)
4 LoRA slots with individual strength controls
PURGE VRAM automatic cleanup
Play/Stop regeneration system (save only finals!)
2D resolution slider (3:4, 4:3, 16:9)
Improved preview/save order (LQ → HQ)

Use when: You want maximum control over text-to-image generation

Extra nodes: MXToolkit, LayerUtility

2. 🎮 Img2Img + ControlNet

ControlNet precision with advanced controls

Key Features:

All Txt2Img features PLUS:
ControlNet Union (Canny, Depth, Pose, HED, MLSD)
Megapixel scaling (auto aspect ratio)
ControlNet strength slider
Input image guidance

Use when: You need precise control with reference images

Extra downloads: ControlNet Union file (~2.5GB) Save in: ComfyUI/models/model_patches/

3. ✨ Img2Img + ControlNet + DETECTION

Professional pipeline with selective detailing

Key Features:

Everything from Img2Img PLUS:
SAM2 auto-segmentation (face, hands, details)
SEGS selective detailer
Model upscaler (4x to 10K resolution)
Grain addition for photographic look
Smart resize to 2K/4K (reasonable file size)
Multiple PURGE VRAM nodes

Use when: You need maximum quality for final outputs

Extra nodes: Impact Pack, SAM2 Extra downloads: Upscaler model, SAM2 model

🆚 Quick Comparison:

|---------|---------|---------|----------|

| Input Image | ❌ | ✅ | ✅ |

| ControlNet | ❌ | ✅ | ✅ |

| SEGS Detailer | ❌ | ❌ | ✅ |

🎯 When to Use Which:

Choose Txt2Img when:

✅ Pure text-to-image generation
✅ Quick iterations and testing
✅ Multiple LoRAs experimentation
✅ Don't need reference images

Choose Img2Img + ControlNet when:

✅ Have reference/input image
✅ Need pose/composition control
✅ Sketch-to-photo conversion
✅ Architectural work
✅ Want guided generation

Choose Detailer when:

✅ Creating portfolio pieces
✅ Professional/commercial work
✅ Need perfect face/hand details
✅ Want photorealistic texture
✅ Maximum quality required
✅ Don't mind longer processing

✨ Shared Features (All 3):

🎛️ Interactive Sliders:

CFG, Steps, Denoise
LoRA strengths (4 slots)
Upscale parameters
ControlNet strength (Img2Img variants)

🔄 Play/Stop System:

Green PLAY = Generate/regenerate
Purple SAVE = Save final only
No cluttered saves folder!

🧹 PURGE VRAM:

Automatic cleanup after generation
Prevents memory buildup
Better performance on all GPUs

📦 Multi-LoRA:

4 LoRA slots
Individual strength sliders
Easy on/off (set to 0.0)

📸 Metadata:

Auto-saved to images
Easy CivitAI uploads

📥 Downloads:

Main Model:
Z-Image-Turbo-AIO FP8/BF16

ControlNet Union (for Img2Img variants):
HuggingFace Download
⚠️ Save in: ComfyUI/models/model_patches/

Test Online:
TensorArt (FP8)

🎯 Required Custom Nodes:

All Workflows:

MXToolkit - Sliders & controls
rgthree-comfy - LoRA stack
LayerUtility - PURGE VRAM
comfyui_image_metadata_extension

Img2Img + ControlNet:

comfyui_controlnet_aux - Preprocessors
⚠️ ComfyUI 3.77+ required!

Detailer:

Impact Pack - SEGS detailer
SAM2 - Segmentation
⚠️ ComfyUI 3.77+ required!

⚙️ Settings (All Workflows):

Steps: 9 (slider adjustable)
CFG: 1.0 (slider adjustable)
Sampler: res_multistep or euler_ancestral
Scheduler: simple or beta
NO negative prompts needed

💡 Pro Tips:

Slider Workflow:

Start with defaults, adjust as needed
Set LoRA to 0.0 to disable
Use PLAY to test variations
Only SAVE final results

ControlNet Strength:

0.3-0.5 = Subtle guidance
0.6-0.8 = Balanced (recommended)
0.9-1.0 = Strong control

Detailer:

Best with 1024px+ input
Let SAM2 auto-detect regions
Grain at 10K = most natural
Downscale to 2K/4K recommended

PURGE VRAM:

Runs automatically
Helps weaker GPUs
Prevents memory issues

🎨 Example Workflow:

Quick Test (Txt2Img):

Load 1-2 LoRAs → 2. Write prompt → 3. PLAY → 4. Adjust sliders → 5. PLAY again → 6. SAVE

Controlled Gen (Img2Img):

Upload reference → 2. Choose preprocessor → 3. Load LoRAs → 4. Write prompt → 5. Adjust strength → 6. PLAY → 7. SAVE

Final Polish (Detailer):

Upload input → 2. Set up ControlNet → 3. Load LoRAs → 4. Write prompt → 5. PLAY (wait 30-60s) → 6. SAVE 2K/4K

❓ FAQ:

Q: Which workflow should I start with?
A: Txt2Img for learning, Img2Img for control, Detailer for finals.

Q: Do I need all custom nodes for all workflows?
A: No - each workflow lists its specific requirements.

Q: What's MXToolkit?
A: Provides the slider interface. Makes adjustments easier.

Q: Why PURGE VRAM?
A: Cleans memory after generation. Especially helpful on 8GB cards.

Q: Detailer too slow?
A: Yes, it's intensive. Use only for final images, not testing.

Q: Can I use original workflows instead?
A: Yes! These are advanced versions. Original workflows still work great.

Q: MXToolkit not working with ComfyUI 2.0?
A: Disable Node 2.0 interface for now. MXToolkit compatibility coming.

🙏 Credits:

Advanced Workflows: sphiratrioth666
Original Workflows: SeeSeeLP
Base Model: Tongyi Lab (Alibaba Group) - Z-Image-Turbo
License: Apache 2.0

Big thanks to sphiratrioth666 for the amazing enhancements! 🎉

📊 System Requirements:

Minimum:

VRAM: 8GB (all workflows tested on RTX 4060)
RAM: 16GB (32GB recommended for Detailer)
ComfyUI: 3.77+ (for ControlNet/Detailer)

Detailer additionally needs:

More processing time (~30-60s)
SAM2 and upscaler models
Patience! 😄

Updated: December 2025
Compatible: Z-Image-Turbo-AIO FP8 & BF16
Tested: RTX 4060 8GB, RTX 5090

"I upgraded your Z-Image workflow by a lot" - sphiratrioth666

Try all three workflows and find your perfect setup! 🚀

此模型生成的图像

排序

模型类型	工作流
基础模型	ZImageTurbo
发布时间	2025-12-12