Z-Image Outpainting and Object Removal: ComfyUI Editing Workflow

Core philosophy: "Paint" where change is needed, leave the rest untouched.

In AI image generation workflows, Z-Image excels not only at generating images from scratch but also at precise editing of existing images. Whether it's expanding e-commerce product shots, extending backgrounds, removing unwanted objects, or repairing details, Z-Image's Outpainting and Inpainting engine can achieve professional-grade retouching results with minimal effort.

This article systematically covers Z-Image's outpainting and object removal workflows in ComfyUI, from core concepts to practical techniques, helping you more than double your editing efficiency.

---

1. Why Outpainting Matters

In real-world commercial scenarios, "generating a perfect image" is often just the first step. More often, you need to deal with:

• Product image dimensions don't match platform requirements: The platform requires a 1:1 square, but you only have a 3:4 vertical image
• Composition needs adjustment: The subject is off-center, requiring canvas expansion and recropping
• Unwanted elements in the frame: Power cables, watermarks, and random passersby need seamless removal
• Background isn't "clean" enough: A white background isn't white enough, requiring extension into a seamless white backdrop

Fixing these issues in traditional Photoshop requires skilled operators working on each image individually. With Z-Image's Outpainting/Inpainting engine, combined with proper prompts and masking strategies, you can compress batch editing time for 9 product images from 40 minutes down to 18 minutes.

This is why mastering these editing skills is more important than simply knowing how to "generate."

---

2. Three Editing Modes Explained

Z-Image's editing capabilities are divided into three modes. Understanding their differences is a prerequisite for efficient workflows.

Inpainting (Local Redraw)

Replaces or modifies specific areas within an existing image. For example:

• Changing clothing color
• Altering facial expressions
• Repairing damaged textures

Outpainting (Canvas Expansion)

Expands the canvas beyond the image boundaries, with AI automatically filling in the extended content. For example:

• Expanding a vertical photo into horizontal format
• Adding more background space to product images
• Generating wider panoramic scenes

Object Removal

A focused variant of Inpainting, specifically designed to remove unwanted objects and intelligently fill in the background. For example:

• Removing power lines from photos
• Clearing people from the background
• Eliminating watermarks and tags

Comparison of the Three Modes

Feature	Inpainting (Local Redraw)	Outpainting (Canvas Expansion)	Object Removal
Target Area	Inside the image	Outside image boundaries	Inside the image
Mask Scope	Large or small, customizable as needed	Usually in edge extension areas	Precisely wraps the object to be removed
Prompt Focus	Describe replacement content	Describe extension direction and style	Describe the background to "preserve"
Typical Scenarios	Outfit change, face retouching, color change	Canvas expansion, panoramic extension	Line removal, watermark removal, clutter cleanup
Iterations	Usually 1–2	Multiple small steps	Usually 1 pass, 2 for complex objects
Mask Precision Requirement	Medium	Low (edge areas)	High (needs precise wrapping)

---

3. Outpainting Canvas Expansion: Step-by-Step Workflow

The success rate of Outpainting depends on whether you follow the principle of small steps, clear direction.

Step 1: Determine Target Aspect Ratio

In ComfyUI, first clarify your final desired dimensions.

• E-commerce white background: 1:1 (1024×1024)
• Horizontal banner: 16:9 (1920×1080)
• Vertical social media: 4:5 (1024×1280)

💡 Recommendation: Use the target dimensions as a starting point, and work backward to calculate the pixel expansion needed for each side.

Step 2: Expand One Side at a Time

This is the most critical step. Do not expand all four sides at once—instead, proceed side by side:

1. Expand left → Generate → Confirm result
2. Expand right on the result → Generate → Confirm result
3. Expand top → Generate → Confirm result
4. Finally, expand bottom → Generate → Confirm result

Keep each expansion amount within 10%–20% (relative to the current canvas width/height) to avoid the AI generating too much unknown area at once, which can cause style drift.

Step 3: Use Directional Prompts

Provide clear visual guidance for each side's expansion:

Expand left 15%: continue wood floor texture, maintain natural light gradient from left to right.

Expand top 10%: white seamless background, consistent color temperature with existing background, retain continuation of mug bottom shadow at the base.

Step 4: Check Seams and Tone

After each expansion, check:

• Whether the boundary between old and new areas is natural (no hard edges, no color mismatch)
• Whether lighting direction is consistent
• Whether textures are continuous (e.g., wood grain, fabric texture)

If issues are found, reduce the expansion amount and regenerate, rather than attempting additional patching.

---

4. Object Removal Strategies

The core philosophy of Object Removal is counterintuitive: Don't just tell the AI what to remove—tell it what to preserve.

Strategy 1: Describe the Texture to "Restore"

❌ Wrong approach: Remove power cable
✅ Correct approach: Remove power cable: restore wood grain and keep natural window reflection: lighting from left.

Explicitly naming the type of texture to restore (wood grain, fabric, wall, skin) enables the AI to generate accurately matching fill content.

Strategy 2: Respect Shadow Relationships

If the object to be removed casts a shadow, you need to decide after removal:

• Shadow belongs to the object → Remove it together, restore the occluded background
• Shadow belongs to another object → Preserve the shadow, only remove the object itself

Specify in the prompt: respect existing shadows or remove object and its shadow.

Strategy 3: Use Irregular Masks

Don't use rectangular bounding boxes to select the object to remove. A mask that precisely follows the object's contour significantly reduces the AI's guesswork area and improves fill accuracy.

• The mask should extend slightly beyond the object edge (1–2px), including a small amount of contextual information
• Avoid masks hugging high-contrast edges, as this can result in harsh blending

Strategy 4: Remove Complex Objects in Steps

For structurally complex objects (e.g., people, animals), you can:

1. First pass: remove the main body
2. Second pass: fix remaining shadows or background imperfections

---

5. Masking Best Practices

Mask quality directly determines editing results—its importance is no less than that of prompts.

Brush Softness Matters More Than Size

⚠️ This is the most common misconception. Many people spend time adjusting mask size while neglecting brush softness (Feather).

Recommended settings:

• Feather: 3–8px (adjust based on image resolution)
  • Low-resolution images (≤512px): 3–5px
  • High-resolution images (1024px+): 5–8px

Proper feathering gives the mask edge a natural transition zone, allowing the AI to perform gradient blending within this area and avoid hard edges or color patches.

Five Golden Rules

Rule	Description
Work at output resolution	Don't draw masks on thumbnails. Always operate at final output resolution to ensure precision
Avoid high-contrast edges	If the mask edge crosses an area of strong light-dark contrast (e.g., the boundary between a white wall and dark furniture), adjust the mask position so the transition zone falls within the same texture region
Include a 1–2px context slice	The mask should slightly exceed the target area, including a few surrounding pixels as style reference for the AI
Save masks for reuse	When processing in batches, save commonly used masks as PNG grayscale images and load them directly later, avoiding repeated drawing
Brush softness takes priority over size	Adjust softness first, then size. If softness is insufficient, enlarging the brush won't fix it

---

6. Prompt Engineering for Editing Scenarios

Prompts in Z-Image's editing mode differ fundamentally from those in generation mode.

Core Principle: One Sentence, Clear Constraints

Editing prompts don't need elegant rhetoric—they need precise instructions. Recommended format:

[Action target] : [What to restore/preserve] : [Environmental constraints]

Practical Prompt Templates

Object Removal

Remove power cable: restore wood grain and keep natural window reflection: lighting from left.

Background Extension

Extend white seamless background 15% on the right: match gradient and soft shadow under mug.

Chinese Prompts

Remove power lines on the right: restore white wall texture, preserve natural light spots on the wall, lighting from upper left.

Extend white seamless background 15% to the right: match background gradient transition, preserve soft shadow beneath the mug.

Fixing Plastic-Look Results

AI-generated fill areas can sometimes appear overly smooth, resulting in a "plastic look." Appending the following phrases at the end of your prompt can effectively mitigate this:

• retain subtle imperfections
• slight surface irregularities

For Chinese prompts: preserve natural texture details, do not over-smooth

Prompt Pitfall Checklist

Problem	Cause	Fix
Fill content doesn't match style	Missing style constraints	Append match surrounding style and color temperature
Area too smooth	Missing texture constraints	Append retain subtle imperfections
Color shift	Missing color temperature constraints	Append match color temperature of original image
Extra objects generated	Prompt too broad	Narrow the description scope, focus on the texture to restore

---

7. Common Issues and Troubleshooting

Problem	Cause	Solution
Edge halo (Halo)	Brush hardness too high, mask edge too sharp	Reduce brush hardness, increase Feather to 5–8px
Blurred/smeared texture	Mask area too large, AI fill exceeds texture repeat cycle	Reduce mask area, process in multiple small areas
Repeating patterns/mirroring	Single expansion too large, AI duplicates existing texture	Reduce per-step expansion pixel amount (≤15%), proceed in steps
Color shift/white balance mismatch	Color temperature mismatch between old and new areas	Append match color temperature in prompt, or manually fine-tune output white balance
Object remnants/incomplete removal	Mask doesn't fully cover the target object	Expand mask area, ensure the entire object is within the mask
Background incorrectly modified	Mask area exceeds target region	Shrink the mask, precisely fit the target object contour
Visible seams	Expansion too large or insufficient directional guidance	Reduce expansion amount, add directional descriptive terms

---

8. Practical Case Studies

Case 1: Expanding E-commerce Product Image to Square

Scenario: A 3:4 vertical product photo (800×1067) needs to be expanded to 1:1 (1067×1067) for an e-commerce platform.

Steps:

1. Calculate expansion amount: left side needs (1067–800)/2 ≈ 134px (~17%)
2. Expand left:
   • Mask: left 150px area, Feather 5px
   • Prompt: Extend white seamless background on the left: match gradient and soft shadow under product.
3. Expand right:
   • Mask: right 150px area, Feather 5px
   • Prompt: Extend white seamless background on the right: match gradient and soft shadow under product.
4. Check seams, confirm consistent tone.

Time: ~2 minutes/image (including generation time)

---

Case 2: Removing Power Cable from Product Photo

Scenario: A white-background product image with a black power cable visible behind the product, needing seamless removal.

Steps:

1. Precisely trace the power cable mask (fitting the cable contour, with 2px outward expansion)
2. Set Feather to 4px
3. Prompt: Remove black power cable: restore white seamless background, maintain shadow consistency: lighting from above.
4. After generation, check the cable area and confirm natural background texture.

Time: ~1 minute/image

---

Case 3: Batch Editing Efficiency Comparison

Scenario: 9 product photos, each requiring background clutter removal + expansion to square.

Method	Time	Result Consistency
Photoshop manual retouching	~40 minutes	Depends on operator skill level
Z-Image semi-automated workflow	~18 minutes	High (templated prompts)
Z-Image + mask reuse	~12 minutes	Highest

💡 Efficiency tip: Save commonly used masks as templates, pair with fixed prompts, and build a reusable "retouching recipe."

---

Summary

Z-Image's Outpainting and Object Removal features essentially automate the most time-consuming "retouching" phase of traditional image editing. Mastering the following key points can significantly improve your editing quality:

1. Small-step expansion: Expand 10–15% per step, proceed side by side, avoid one-step completion
2. Precise masking: Feather 3–8px is the golden range; softness matters more than size
3. Clear prompts: One sentence that says what to do, what to restore, and the environment
4. Tell the AI what to preserve: More important than describing what to remove
5. Prevent plastic look: Append retain subtle imperfections

These techniques are applicable not only to Z-Image workflows in ComfyUI, but their core philosophy—precise mask + clear instruction + small-step iteration—also applies to other AI image editing tools.

---

📌 Coming up next: Z-Image ControlNet Precise Image Control: A Complete Workflow from Line Art to Photorealistic Rendering

---

This article is based on Z-Image v2.0 tested in ComfyUI 1.6+. Different versions may have node differences—please refer to your actual version.