Setup Page

The Setup page is where you create your base image. This is the foundation for all your variations.

Page Layout

• Canvas (center) - Displays your generated or uploaded image
• Inspector panel (right) - Configuration options
• Thumbnail stack (bottom-right) - Switch between base images
• JSON button (bottom-left) - Advanced configuration editor

Creating Images

Writing a Prompt

Enter your image description in the Image Prompt field:

A cute robot mascot with large expressive eyes, metallic silver body,
small antenna on top, digital art style, centered, white background

Prompt tips:

• Be specific about style, colors, and composition
• Include "centered" for easier trait detection
• Use "white background" or "solid background" for clean extraction
• Describe distinct parts you'll want to vary

Choosing a Model

Model	Provider	Speed	Max Size
Gemini 2.5 Flash Image	Google	Fast	1024px
Gemini 3.0 Pro Image	Google	Slower	4096px
FLUX.2 [pro]	Black Forest Labs	Moderate	2048px
GPT Image 1	OpenAI	Moderate	1024px
GPT Image 1.5	OpenAI	Moderate	1024px
Imagen 4	Google	Moderate	2048px
Imagen 4 Ultra	Google	Slower	2048px

Recommendation: Use Gemini Flash for iteration, switch to a premium model for final images.

Costs vary by model and settings. Check the estimate badge on the Generate button.

Aspect Ratio

Choose from 10 aspect ratios:

Ratio	Use Case
1:1	Avatars, icons, square art
3:2, 2:3	Photography style
4:3, 3:4	Traditional art
16:9, 9:16	Widescreen, mobile
21:9	Ultra-wide

Image Size

Available on models that support multiple resolutions (Gemini Pro, FLUX, Imagen, Seedream):

Size	Resolution
1K	1024px
2K	2048px
4K	4096px (Gemini Pro only)

Quality Level

Available on GPT Image models only. Controls output fidelity:

• Low — Fastest, lowest cost
• Medium — Balanced quality and cost
• High — Best quality, highest cost
• Auto — Model chooses optimal quality (GPT Image 1.5 only)

Generating

Click the Generate button in the footer. The image appears in the canvas when complete.

Batch Generation

Generate multiple base images in a single operation:

1. Set the number of images to generate (1-5)
2. Click Generate — images auto-assign to empty slots
3. A single Mana deduction covers the entire batch

The cost scales with the number of images and is shown on the button before you generate.

Managing Base Images

Multiple Images

You can create up to 5 base images per project using the thumbnail stack:

• Click + to add a new slot
• Click a thumbnail to select it
• Each slot can hold one image

Selecting the Base Image

The starred image is your "base image" - the one used for the rest of the pipeline (Detection, Variations, Combine).

To change the base image:

1. Click the star icon on any image
2. If you have existing masks/variations, you'll be prompted to delete them

Changing Base Image

Switching your base image after creating masks requires regenerating all downstream content. VariantLab will warn you before deleting existing work.

Regenerating

Click the regenerate icon (circular arrows) on any image to replace it with a new generation using the current prompt and settings.

Deleting

Click the trash icon to delete an image from a slot.

Context Menu

Right-click any base image thumbnail for quick actions:

• Use as Reference — Set the image as a reference for generation (available on supported models)
• Revise — Edit the image with custom instructions using an AI edit model
• Delete — Remove the image from its slot

Uploading Images

You can upload your own images instead of generating:

1. Click the Upload button, or
2. Drag and drop an image onto the canvas

Uploaded images work the same as generated ones in the pipeline.

Background Removal

Remove backgrounds after generation:

1. Enable Remove Background toggle before generating, or
2. Click the magic wand icon on an existing image

Background Removal Models

Model	Best For
U2Net Fast	General purpose, fast
ISNet General	Clip art, digital art
U2Net Pro	High quality general
U2Net Cloth	Clothing and textiles
U2Net Human	People and portraits
Silueta	High quality general
ISNet Anime	Anime and manga

Restoring Original

If you've removed the background, click the undo icon to restore the original image.

JSON Editor

Click the JSON button to access the full project configuration:

{
  "prompt": "Your image prompt...",
  "config": {
    "model": "gemini-2.5-flash-image",
    "aspectRatio": "1:1",
    "imageSize": "1K",
    "remove_bg": false,
    "bg_removal_model": "u2net"
  },
  "analysis": {
    "traits": ["eyes", "antenna"],
    "layer_order": ["eyes", "antenna"],
    "detection_prompts": {}
  },
  "variations": []
}

Edit and save to update all settings at once.

Tips

1. Generate multiple options - Create 3-5 base images and pick the best
2. Check for clear regions - Ensure traits are visually distinct
3. Use consistent style - All variations should match your base image's style
4. Start with simple subjects - Complex images are harder to mask
5. Save your prompt - Good prompts can be reused in other projects