Z-Image Edit Image Editing Workflow: Inpainting/Outpainting/Object Removal Complete Guide
Published: 2026-05-13
Author: Z-Image Tech Team
Tags: Z-Image Edit, Inpainting, Outpainting, Object Removal, Image Editing, ComfyUI
Overview
Z-Image Edit is a variant model based on Z-Image Base, specifically fine-tuned for image editing tasks. Unlike general-purpose text-to-image models, Z-Image Edit understands the semantics of editing instructions, enabling precise execution of Inpainting, Outpainting, and Object Removal operations.
This guide comprehensively covers the three core editing modes of Z-Image Edit, helping you build efficient, professional-grade image editing workflows.
Model Selection: Z-Image Edit vs Z-Image Base
Key Differences
| Feature | Z-Image Base | Z-Image Edit |
|---|---|---|
| Training Objective | Text-to-image generation | Image editing instruction following |
| Inpainting | Requires additional training | Native support |
| Outpainting | Requires additional training | Native support |
| Object Removal | Requires additional training | Native support |
| Edit Instruction Understanding | Weak | Strong |
When to Use Z-Image Edit
- Making local modifications to existing images
- Extending image frames (adding background, adjusting composition)
- Removing unwanted elements
- Maintaining style consistency with the original image
When to Use Z-Image Base
- Generating new images from scratch
- LoRA training (Edit model is not suitable for training)
- Pure generation scenarios without editing needs
Mode 1: Inpainting (Local Redrawing)
Inpainting allows you to specify a region via a mask and regenerate that area while keeping the masked-out areas unchanged.
Core Workflow
Original Image + Mask → Z-Image Edit → Edited Image
Use Cases
- Face Repair: Improve expressions, change clothing, adjust hairstyles
- Product Replacement: Change product backgrounds, add/modify product elements
- Local Style Transfer: Apply different styles to specific areas
- Detail Enhancement: Zoom and regenerate blurry areas
ComfyUI Node Configuration
1. LoadImage — Load original image
2. Mask — Create/load mask area
3. VAEEncodeForInpaint — Encode to inpaint latent
4. CLIPTextEncode — Edit prompt
5. KSampler (Edit Mode) — Execute edit sampling
6. VAEDecode — Decode result
7. SaveImage — Save
Key Parameters
| Parameter | Recommended | Description |
|---|---|---|
| Denoise | 0.7-0.9 | High denoise allows more change |
| Steps | 20-30 | Edit model needs more steps |
| CFG | 5.0-7.0 | Higher CFG ensures instruction following |
| Mask Blur | 4-8px | Soft mask edge transition |
Mask Creation Tips
Method 1: Built-in ComfyUI Mask Tools
- Use
DrawMasknode for manual drawing - Supports rectangle, circle, free brush
- Adjustable mask feather radius
Method 2: SAM2 Automatic Segmentation
- Use Segment Anything 2 to auto-detect objects
- Auto-mask from text descriptions
- Ideal for precise segmentation of complex shapes
Method 3: External Tools
- Photoshop/Photopea for mask creation
- Rembg for auto-background removal
- Import into ComfyUI for use
Mode 2: Outpainting (Frame Extension)
Outpainting extends the image "outward" from the edges, generating new content beyond the original image boundaries.
Core Workflow
Original Image + Extension Direction → Z-Image Edit → Extended Image
Use Cases
- Composition Adjustment: Extend portrait to landscape or vice versa
- Background Extension: Add more background space to images
- Social Media Adaptation: Extend square images to 16:9 video covers
- Creative Extension: Add new elements in specific directions
ComfyUI Node Configuration
1. LoadImage — Load original image
2. ImageScale — Adjust original image position (if centering needed)
3. ImagePadForOutpaint — Fill extension area
4. Mask — Mark extension area (blank area = mask)
5. VAEEncodeForInpaint — Encode
6. CLIPTextEncode — Describe extension content
7. KSampler (Edit Mode) — Execute extension
8. VAEDecode — Decode
9. SaveImage — Save
Key Parameters
| Parameter | Recommended | Description |
|---|---|---|
| Extension Width | 25%-50% of original width | Gradual extension works best |
| Denoise | 0.8-0.9 | High denoise allows full creativity |
| Steps | 20-30 | Edit model needs more steps |
| CFG | 5.0-7.0 | Ensure natural edge transitions |
Multi-Step Extension Tips
For large extensions, perform in multiple steps:
Step 1: Extend 25% → Step 2: Extend another 25% → Step 3: Extend another 25%
Advantages:
- Review results at each step, adjust as needed
- Avoid quality degradation from single large extension
- More natural edge transitions
Mode 3: Object Removal
Object removal is essentially a special form of Inpainting — mask the object to remove, then let the model fill the area with surrounding content.
Core Workflow
Original Image + Object Mask → Z-Image Edit (empty prompt or background prompt) → Cleaned Image
Use Cases
- Watermark/Logo Removal: Remove copyright marks from images
- Clutter Cleanup: Remove power lines, passersby, and distracting elements
- Product Retouching: Remove blemishes from product photos
- Privacy Protection: Remove sensitive information or faces
Key Techniques
Prompt Strategy:
- Method 1: Empty prompt — Let the model fill in the background autonomously
- Method 2: Background description — Describe what the removed area should be
- Example: Remove power lines → Prompt "clear blue sky, no wires"
- Example: Remove passerby → Prompt "empty park path, green grass"
Mask Precision:
- Mask must precisely cover the object to remove
- Add 2-4px feathering to avoid hard edges
- For transparent/semi-transparent objects (like wires), make the mask slightly larger
Handling Complex Removals
Scenario: Removing an object that spans a large area (like a person across the frame)
Strategy:
- Remove the main subject first
- Use lower denoise (0.6-0.7) to maintain background consistency
- If results aren't ideal, perform a second refinement round
- Use more precise prompts in the second round
Comprehensive Workflow: Mixed Editing
In real scenarios, you often need to combine multiple editing modes. Here's a typical mixed editing workflow:
Case Study: Product Photo Retouching
Step 1: Object Removal
- Remove clutter from the background
- Prompt: clean white background, studio lighting
Step 2: Inpainting
- Fix surface blemishes on the product
- Prompt: pristine condition, clean surface, no scratches
Step 3: Outpainting
- Extend background, add product display space
- Prompt: minimalist studio background, soft lighting
Advanced Techniques
1. Edge Blending Optimization
The boundary between edited areas and the original often shows unnatural transitions. Solutions:
- Mask Feathering: Set 4-8px feather radius
- Block Editing: Split large edits into multiple small blocks
- Iterative Refinement: First pass rough edit + second pass fine tuning
2. Style Consistency
Key to ensuring edited areas match the original style:
- Reuse Prompts: Include original image style description in edit prompts
- IP-Adapter Reference: Use original image as IP-Adapter reference input
- Low Denoise: Use lower denoise (0.5-0.6) to reduce style drift
3. Batch Editing Automation
Use ComfyUI's batch processing capabilities:
FolderInput(original images) → SAM2(auto mask) → KSampler(Edit) → SaveImage
Common Issues and Solutions
Issue 1: Edited Area Doesn't Match Edges
Cause: Insufficient mask feathering or denoise too high.
Solution:
- Increase mask feathering to 6-8px
- Lower denoise to 0.7
- Use second pass with low denoise (0.3) to blend edges
Issue 2: Style Inconsistency After Outpainting
Cause: Model lacks sufficient context to understand original image style.
Solution:
- Describe original image style in detail in prompts
- Use smaller extension steps (25%)
- Consider using IP-Adapter for style preservation
Issue 3: Traces Remain After Object Removal
Cause: Imprecise mask or model can't understand background context.
Solution:
- Precisely adjust mask boundaries
- Provide clear background description prompts
- Remove in stages, handling smaller areas each time
Performance Benchmarks
| Edit Type | Input Resolution | Steps | Processing Time (RTX 3090) | VRAM |
|---|---|---|---|---|
| Inpainting | 1024×1024 | 20 | ~6s | ~6GB |
| Outpainting (25%) | 1024×1280 | 25 | ~8s | ~7GB |
| Object Removal | 1024×1024 | 20 | ~6s | ~6GB |
| Mixed Editing (3 steps) | 1024→1280 | 20×3 | ~20s | ~8GB |
Summary
Z-Image Edit is a powerful image editing tool that, compared to general generation models:
- Native support for Inpainting, Outpainting, and Object Removal
- Better instruction understanding, reducing trial and error
- Improved style consistency, edited areas blend naturally with originals
- Simplified workflow, professional editing without additional training
Master the three core modes of Z-Image Edit and you can effortlessly handle everything from simple touch-ups to professional-grade image retouching.