Skip to main content

Background Removal

VariantLab can automatically remove backgrounds from your images using AI-powered segmentation models.

How It Works

Background removal uses machine learning to:

  1. Detect the foreground subject
  2. Create a mask separating subject from background
  3. Remove background pixels
  4. Replace with transparency

The result is a PNG with a transparent background.

When to Use

Before Generation

Enable Remove Background in Setup settings:

  • Background removed automatically after generation
  • Original with background is preserved
  • Can restore original later

After Generation

Click the magic wand icon on any base image:

  • Removes background from existing image
  • Uses currently selected model
  • Original is preserved

Removal Models

VariantLab offers 7 background removal models:

U2Net Fast

  • Speed: Fastest
  • Best for: General purpose, quick previews
  • Quality: Good

ISNet General

  • Speed: Fast
  • Best for: Digital art, clip art, illustrations
  • Quality: Very good for flat styles

U2Net Pro

  • Speed: Medium
  • Best for: High quality general segmentation
  • Quality: Excellent

U2Net Cloth

  • Speed: Medium
  • Best for: Clothing, textiles, fashion
  • Quality: Specialized

U2Net Human

  • Speed: Medium
  • Best for: People, portraits, characters
  • Quality: Specialized for humans

Silueta

  • Speed: Medium
  • Best for: High quality general segmentation
  • Quality: Excellent

ISNet Anime

  • Speed: Fast
  • Best for: Anime, manga, cartoon styles
  • Quality: Specialized for anime

Choosing a Model

Image TypeRecommended Model
General/MixedU2Net Fast or Silueta
Digital artISNet General
Anime/CartoonISNet Anime
People/CharactersU2Net Human
Clothing/FashionU2Net Cloth
High quality neededU2Net Pro or Silueta
Quick previewU2Net Fast

Tips for Better Results

Image Composition

  • Clear subject - Subject should be distinct from background
  • Contrast - Different colors help separation
  • Clean edges - Avoid subjects that blend into background

Prompt Tips

When generating images, include:

...centered, solid white background, full body visible

This creates images that are easier to segment.

Model Selection

  • Start with U2Net Fast for quick tests
  • Switch to specialized models if needed
  • Use Silueta or U2Net Pro for final quality

Restoring Original

If you've removed the background:

  1. Click the undo icon on the image
  2. The original (with background) is restored
  3. You can remove the background again anytime

The original is always preserved internally.

Background Removal vs. Trait Masks

These serve different purposes:

FeatureBackground RemovalTrait Masks
PurposeRemove image backgroundDefine trait regions
ScopeEntire backgroundSpecific traits
EditingAutomatic onlyManual refinement
Use caseClean base imagesLayer extraction

You might use both:

  1. Remove background from base image
  2. Create trait masks for variations
  3. Extract layers with clean edges

Troubleshooting

Background not fully removed

  • Try a different model
  • Image may have low contrast
  • Some areas may need manual work in editing software

Subject partially removed

  • Model may have misidentified foreground
  • Try a different model
  • Ensure subject is clearly distinct

Edges look rough

  • Try Silueta or U2Net Pro for better edges
  • May need post-processing in editing software
  • Consider using trait masks for more control