Inpainting Workflow

Fix specific areas of an image.

Concept

Inpainting regenerates masked areas while preserving the rest.

Workflow

[LoadImage] → [VAEEncode (with mask)] → [KSampler] → [VAEDecode] → [SaveImage]
               ↑ CLIP Text Encode

Step-by-Step

Step 1: Load Image

Node: LoadImage
- Load your image
- If using mask, load RGBA image

Step 2: Create Mask

Option A: RGBA Image

Load PNG with transparency:

LoadImage (RGBA)
    ↓ IMAGE → VAEEncode
    ↓ MASK → Use as mask

Option B: Internal Mask Editor

In ComfyUI-Impact-Pack or similar:

1. Right-click → Mask Editor
2. Paint white areas to regenerate
3. Black areas stay unchanged

Step 3: Encode

Node: VAEEncode (Inpainting or Masked)
- Connect image
- Connect mask (if separate)

Step 4: Set Parameters

Node: KSampler
Denoise: 0.6-0.9

Higher = more regeneration
Lower = more faithful to original

Step 5: Generate

Click "Queue Prompt"

Parameters

Denoise Strength

Value	Effect
0.3-0.5	Subtle fix
0.6-0.7	Moderate change
0.8-0.9	Major regeneration

Mask Settings

Setting	Value	Description
Feather	5-10px	Soft edges
Invert	Off	White = regenerate

Use Cases

Fix Face

Input: Image with bad face
Mask: Cover face area
Prompt: "beautiful face, detailed features"
Denoise: 0.7-0.8

Remove Object

Input: Image with unwanted object
Mask: Cover object
Prompt: "clean background, seamless"
Denoise: 0.8-0.9

Add Object

Input: Image
Mask: Area to add to
Prompt: "add a red flower vase"
Denoise: 0.7-0.8

Tips

1. Feather edges: Smooth transition for natural look
2. Cover context: Include surrounding area in mask
3. Match style: Use similar prompts to original

Common Issues

Issue	Solution
Visible seams	Increase feather
Style mismatch	Adjust prompt
Wrong content	Use lower denoise

Next Steps

• Outpainting - Extend image
• Upscaling - Increase resolution