🚀 Z-Image AIO | Official Workflows

Turbo (8 Steps) & Base (28-50 Steps) • Photorealistic Generation • Bilingual Text • FP8 / FP16 / BF16

⚠️ Requires ComfyUI v0.11.0+ → Download here

📦 Eight Official Workflows Available

All workflows work with FP8 (~10GB), FP16 (~20GB), and BF16 (~20GB) versions!

🆕 Z-Image-Base-AIO Workflow (ZIB-AIO-Base)

Full foundation model with maximum creative control!

The undistilled 6B parameter model for professional work. Features full CFG control (3.0-5.0), negative prompt support, and high output diversity. Ideal base for LoRA training and complex prompt engineering. Includes SeedVR2 upscaler and Seed Variance Enhancer as optional features.

Key Features:

• ✅ Full CFG control (3.0-5.0)

• ✅ Negative prompts supported

• ✅ High output diversity

• ✅ Ideal for LoRA/ControlNet development

• ✅ Optional LoRA loading via LoraManager

• ✅ Optional SeedVR2 upscaling

• ✅ Optional Seed Variance Enhancer

Required Custom Nodes: 5 nodes (see below)

🚀 Standard Workflow v1.0 & v2.0 (ZIT-AIO-v1.0 / ZIT-AIO-v2.0)

Simple text-to-image workflow with improved upscaler and dual sampler options. Features automatic metadata saving and denoise control for upscaling. Perfect for beginners and quick generations. Requires 2 custom nodes.

v2.0 improvements: Enhanced upscaler with denoise control, better sampler options (res_multistep or euler_ancestral), dual scheduler support (simple or beta).

🎮 ControlNet Workflow (ZIT-AIO-Control)

Guided generation with reference images using ControlNet Union (Canny, HED, Depth, Pose, MLSD). Uses megapixel scaling that maintains aspect ratio automatically. Perfect for sketch-to-photo, pose transfer, and precise composition control. Requires ComfyUI 3.77+ and ControlNet Union file.

🎲 Seed Variance Enhancer Workflow (ZIT-AIO-Variance)

Adds diversity to outputs by introducing controlled noise to text embeddings. Compensates for low seed variance - get more varied results with the same prompt. Includes manual seed control for reproducibility. Requires SeedVarianceEnhancer custom node.

🎬 SeedVR2 Video Upscaler Workflow (ZIT-AIO-SeedVR2)

Professional diffusion-based upscaling using DiT (Diffusion Transformer) models. Delivers superior quality with temporal consistency for videos and images. Supports multiple model variants (3B/7B with FP16/FP8/GGUF) and memory optimization options. Requires SeedVR2 custom node.

🌊 Depth Anything V3 ControlNet Workflow (ZIT-AIO-DepthV3)

State-of-the-art depth-guided generation with dual modes: create depth-controlled images OR preview depth as 3D point clouds. Superior multi-view depth consistency compared to traditional methods. Features toggle system for easy mode switching. Requires Depth Anything 3 custom nodes.

🖼️ Z-Image-Turbo-Anime Workflow (ZIT-AIO-Anime)

This workflow includes several small but meaningful adjustments and integrates multiple custom nodes. It features the Seed Variance Enhancer, which helps generate different image variations from the same prompt by increasing effective seed diversity. Additionally, SeedVR2 is included as an alternative upscaling solution.

Z-Image-Turbo-Anime: https://civitai.com/models/2259646/z-image-turbo-anime

📊 Quick Comparison

Turbo Workflows (8-9 Steps, CFG 1.0)

🚀 Standard v1/v2 → Text-to-image → Simple & fast → 2 custom nodes

🎮 ControlNet → Guided generation → 5 control types → 3 nodes + ControlNet file

🎲 Seed Enhancer → Output diversity → More variations → SeedVarianceEnhancer node

🎬 SeedVR2 → Professional upscaling → Diffusion-based → SeedVR2 node + models

🌊 DA3 DepthV3 → Depth-guided + 3D → Dual modes → Depth Anything 3 nodes

🖼️ Anime → Anime style → Custom merged → 5 custom nodes

Base Workflow (28-50 Steps, CFG 3.0-5.0)

🆕 Base-AIO → Full control → CFG + Negative prompts → 5 custom nodes

🔄 Model Versions Available

Z-Image-Turbo-AIO (8 Steps, CFG 1.0)

🟡 FP8-AIO (~10GB) - Recommended for most users

🔵 FP16-AIO (~20GB) - Wide GPU compatibility

🌟 BF16-AIO (~20GB) - Maximum quality

Z-Image-Base-AIO (28-50 Steps, CFG 3.0-5.0) 🆕

🟡 FP8-AIO (~10GB) - Fast, daily use

🔵 FP16-AIO (~20GB) - Wide GPU compatibility (RTX 2000/3000)

🌟 BF16-AIO (~20GB) - Max quality, ideal for LoRA training

All versions work on 8GB VRAM!

🆚 Turbo vs Base - When to Use Which?

Use Turbo when:

⚡ Speed is priority - 8 steps = 3-5 seconds

📸 Production workflows - Consistent high quality

💾 Quick iterations - Rapid prototyping

🎯 Simple prompts - Less complex scenes

Use Base when:

🎨 Creative exploration - Higher diversity across seeds

🔧 LoRA/ControlNet development - Undistilled foundation

📝 Complex prompt engineering - Full CFG control

🚫 Negative prompting needed - Remove unwanted elements

🎯 Maximum control - Fine-tune every aspect

⚙️ Settings by Model Type

Z-Image-Turbo Settings (All Turbo Workflows)

📊 Steps: 8-9 🎚️ CFG: 1.0 (don't change!) 🚫 Negative Prompt: ❌ Not used 🎲 Sampler: res_multistep (sharp) / euler_ancestral (smooth) 📈 Scheduler: simple (clean) / beta (balanced)

Z-Image-Base Settings (Base-AIO Workflow) 🆕

📊 Steps: 28-50 🎚️ CFG: 3.0-5.0 🚫 Negative Prompt: ✅ Full support! 🎲 Sampler: euler ⭐ / dpmpp_2m 📈 Scheduler: normal ⭐ / karras

💡 Prompting Tips

Natural Language Works Best

Professional food photography of artisan breakfast plate. 
Golden poached eggs on sourdough toast, crispy bacon, fresh 
avocado slices. Morning sunlight creating warm glow. Shallow 
depth of field, magazine-quality presentation.

Bilingual Text Rendering

Neon sign reading "COFFEE SHOP" in bright blue letters
Sign with "咖啡店" in elegant gold calligraphy

Important

Turbo:

• ❌ NO negative prompts (model ignores them)

• ✅ Natural language, not tags

• ✅ Detailed (100-300 words)

Base:

• ✅ Negative prompts work great!

• ✅ Natural language, not tags

• ✅ Detailed (100-300 words)

• ✅ Use CFG 3.0-5.0 for control

📥 Downloads

Main Models

Z-Image-Turbo-AIO:

• CivitAI: https://civitai.com/models/2173571?modelVersionId=2448013

• HuggingFace: https://huggingface.co/SeeSee21/Z-Image-Turbo-AIO

Z-Image-Base-AIO: 🆕

• CivitAI FP8: https://civitai.com/models/2173571?modelVersionId=2637423

• CivitAI FP16: https://civitai.com/models/2173571?modelVersionId=2638374

• CivitAI BF16: https://civitai.com/models/2173571?modelVersionId=2638695

Additional Files

• ControlNet Union (for ZIT-AIO-Control): https://huggingface.co/alibaba-pai/Z-Image-Turbo-Fun-Controlnet-Union

• SeedVR2 Models (for SeedVR2 workflows): Auto-downloads on first use

• Depth Anything V3 Models (for ZIT-AIO-DepthV3): Auto-downloads on first use

📦 Custom Nodes

Required for ALL Workflows

rgthree-comfy https://github.com/rgthree/rgthree-comfy

comfyui_image_metadata_extension https://github.com/edelvarden/comfyui_image_metadata_extension

Additional per Workflow

ZIB-AIO-Base (Base Workflow): 🆕

• ComfyUI-Lora-Manager: https://github.com/willmiao/ComfyUI-Lora-Manager

• SeedVarianceEnhancer: https://github.com/ChangeTheConstants/SeedVarianceEnhancer

• ComfyUI-SeedVR2_VideoUpscaler: https://github.com/numz/ComfyUI-SeedVR2_VideoUpscaler

ZIT-AIO-Control:

• comfyui_controlnet_aux: https://github.com/Fannovel16/comfyui_controlnet_aux

ZIT-AIO-Variance:

• SeedVarianceEnhancer: https://github.com/ChangeTheConstants/SeedVarianceEnhancer

ZIT-AIO-SeedVR2:

• ComfyUI-SeedVR2_VideoUpscaler: https://github.com/numz/ComfyUI-SeedVR2_VideoUpscaler

ZIT-AIO-DepthV3:

• Depth Anything 3: https://github.com/ByteDance-Seed/Depth-Anything-3

ZIT-AIO-Anime:

• ComfyUI-Lora-Manager: https://github.com/willmiao/ComfyUI-Lora-Manager

• SeedVarianceEnhancer: https://github.com/ChangeTheConstants/SeedVarianceEnhancer

• ComfyUI-SeedVR2_VideoUpscaler: https://github.com/numz/ComfyUI-SeedVR2_VideoUpscaler

💡 Tip: Use ComfyUI Manager → "Install Missing Custom Nodes" for easy installation!

🎯 Workflow-Specific Details

🆕 Base-AIO (ZIB-AIO-Base)

• Steps: 28-50 (more = better quality)

• CFG: 3.0-5.0 (4.0 recommended start)

• Sampler: euler (sharp) / dpmpp_2m (smooth)

• Scheduler: normal (standard) / karras (refined)

• Negative prompts: ✅ Full support!

• Upscaler: Optional with denoise 0.35

• LoRA: Optional via LoraManager node

• SeedVR2: Optional for AI upscaling

• Seed Variance: Optional for diversity

🚀 Standard v2.0

• Improved upscaler with denoise control (0.4-0.6)

• Dual sampler support

• scale_by parameter for output size

• Perfect for everyday use

🎮 ControlNet

• 5 control types: Canny, HED, Depth, Pose, MLSD

• Megapixel scaling (auto aspect ratio)

• ControlNet strength: 0.6-0.8 recommended

• ⚠️ Save ControlNet in: ComfyUI/models/model_patches/

🎲 Seed Variance Enhancer

• randomize_percent: 50

• strength: 20-30

• noise_insert: 'noise on beginning steps'

• Trade-off: Diversity vs prompt adherence

🎬 SeedVR2

• resolution: 1536 (target for short edge)

• batch_size: MUST be 4n+1 (1, 5, 9, 13, 17, 21...)

• color_correction: 'lab' (recommended)

• Models: 3B (faster) or 7B (higher quality)

🌊 Depth Anything V3

• Two modes: Generate images OR 3D preview

• Models: da3_small/base/large/giant

• 8GB VRAM: Use up to da3_large

• ⚠️ Toggle correctly: Generate mode ≠ 3D Preview mode

❓ FAQ

Q: Which workflow should I use? A: Turbo Standard v2 for quick text-to-image. Base-AIO for full control & LoRA work. Others for specific needs.

Q: FP8, FP16, or BF16? A: FP8 for most users (10GB). FP16 for older GPUs. BF16 for maximum quality (20GB).

Q: Turbo or Base? A: Turbo for speed (8 steps). Base for control & quality (28-50 steps, CFG, negative prompts).

Q: Which sampler is better? A: Turbo: res_multistep (sharp) or euler_ancestral (smooth). Base: euler or dpmpp_2m.

Q: Metadata saved automatically? A: Yes! All workflows save metadata for easy CivitAI uploads.

Q: Works on 8GB VRAM? A: Yes! All workflows tested on RTX 4060 8GB.

Q: Do I need all custom nodes? A: No! Only install nodes for workflows you'll use. Base requirement is 2 nodes (rgthree + metadata).

Q: Can I use LoRAs with Base? A: Yes! Base-AIO includes LoraManager for easy LoRA loading. Turbo doesn't support LoRA training well.

🌟 Key Features

All Workflows

✨ Fast Generation - Turbo: 3-5 sec | Base: 30-60 sec 📦 All-in-One - VAE + Text Encoder integrated 📸 Photorealistic - Professional quality 📖 Bilingual - English & Chinese text rendering 💾 Metadata Auto-Save - Easy sharing 🎯 8GB VRAM Friendly - Accessible to everyone

Base-Specific 🆕

⚙️ Full CFG Control - 3.0-5.0 range 🚫 Negative Prompts - Remove unwanted elements 🎨 High Diversity - More variation across seeds 🔧 LoRA Ready - Ideal for training & using LoRAs

Turbo-Specific

⚡ Ultra-Fast - 8-9 steps only 🎯 Consistent - Same quality every time 💾 Efficient - Lower compute needed

🎨 Perfect For

Base-AIO:

• LoRA training & testing

• Complex compositions

• Fine-tuned control

• Professional projects

• Creative exploration

Turbo Workflows:

• Quick iterations

• Production workflows

• Social media content

• Marketing materials

• Rapid prototyping

Both:

• Product photography

• Architectural visualization

• Food photography

• Portrait photography

• Bilingual content

📐 System Requirements

Minimum:

• VRAM: 8GB

• RAM: 16GB

• ComfyUI: v0.11.0+

Recommended:

• VRAM: 8GB+ (perfect for all workflows)

• RAM: 32GB

• Storage: 50GB+ (for all models)

Tested Hardware:

• RTX 4060 8GB @ 1920×1088

• All FP8, FP16, and BF16 versions work perfectly

🙏 Credits

Original Model: Tongyi Lab (Alibaba Group) T

ext Encoder: Qwen3-4B

ControlNet Union: Alibaba PAI Team

SeedVR2: ByteDance Seed Team

Depth Anything V3: ByteDance Seed Team

Architecture: Single-Stream DiT (6B parameters)

License: Apache 2.0

Workflows: Optimized for ComfyUI with metadata support

Community: Thanks to all testers and contributors!

📊 File Sizes

Main Models:

• Turbo FP8-AIO: ~10GB

• Turbo FP16-AIO: ~20GB

• Turbo BF16-AIO: ~20GB

• Base FP8-AIO: ~10GB

• Base FP16-AIO: ~20GB

• Base BF16-AIO: ~20GB

Additional Files:

• ControlNet Union: ~2.5GB

• SeedVR2 models: 10-20GB (3B-7B variants)

• Depth Anything V3: 80MB-1.15GB (model dependent)

Total (all workflows): ~70GB for complete setup

🎯 Getting Started

1️⃣ Download model (FP8, FP16, or BF16 - Turbo or Base)

2️⃣ Install base custom nodes (rgthree + metadata)

3️⃣ Choose workflow based on your needs

4️⃣ Install workflow-specific nodes if needed

5️⃣ Load workflow into ComfyUI v0.11.0+

6️⃣ Generate!

Updated: January 2026 Tested: RTX 4060 8GB @ 1920×1088 ComfyUI: v0.11.0+ required

Eight powerful workflows for every creative need! 🚀

Turbo for speed | Base for control | Choose what fits your workflow!