Skip to main content

AI Image Generation

VariantLab supports multiple AI models from different providers for image generation. Choose the model that best fits your needs based on quality, speed, resolution, and style.

Available Models

Gemini 2.5 Flash Image

Google's fast, cost-effective image generation model. Great for rapid iteration and exploring prompt ideas.

SpecificationValue
ProviderGoogle
Max Resolution1024px
SpeedFast (seconds)
Aspect Ratios10 options

Best for: Prototyping, testing prompts, high-volume generation, budget-conscious projects.

Gemini 3.0 Pro Image

Google's high-quality generation model with support for up to 4K resolution.

SpecificationValue
ProviderGoogle
Max Resolution4096px
SpeedSlower (10-30 seconds)
Size Options1K, 2K, 4K
Aspect Ratios10 options

Best for: Final production images, print-quality output, projects requiring high resolution and detail.

FLUX.2 [pro]

High-fidelity image generation from Black Forest Labs with a commercial license.

SpecificationValue
ProviderBlack Forest Labs
Max Resolution2048px
SpeedModerate
Size Options1K, 2K
Aspect Ratios10 options

Best for: Commercial projects, photorealistic styles, high-quality generation with licensing clarity.

note

FLUX.2 [pro] is generation-only. Variations use FLUX.2 [pro] Edit automatically.

GPT Image 1

OpenAI's multimodal image generation model with quality level control.

SpecificationValue
ProviderOpenAI
Max Resolution1024px
SpeedModerate
Quality LevelsLow, Medium, High
Aspect Ratios3 options (1:1, 2:3, 3:2)

Best for: Diverse art styles, text rendering in images, OpenAI ecosystem integration.

GPT Image 1.5

OpenAI's latest image model — more cost-effective with an additional "auto" quality option.

SpecificationValue
ProviderOpenAI
Max Resolution1024px
SpeedModerate
Quality LevelsLow, Medium, High, Auto
Aspect Ratios3 options (1:1, 2:3, 3:2)

Best for: Cost-effective OpenAI generation, letting the model choose optimal quality with "auto" mode.

Imagen 4

Google's latest dedicated image generation model with excellent prompt adherence.

SpecificationValue
ProviderGoogle
Max Resolution2048px
SpeedModerate
Size Options1K, 2K
Aspect Ratios5 options

Best for: High-quality generation, photorealistic output, strong prompt following.

note

Imagen 4 is generation-only. Variations automatically use a compatible Gemini model.

Imagen 4 Ultra

The highest-quality Imagen model with enhanced detail and output quality.

SpecificationValue
ProviderGoogle
Max Resolution2048px
SpeedSlower
Size Options1K, 2K
Aspect Ratios5 options

Best for: Maximum quality output, detailed illustrations, premium production images.

note

Imagen 4 Ultra is generation-only. Variations automatically use a compatible Gemini model.

Seedream 4.5

ByteDance's unified generation and editing model.

SpecificationValue
ProviderByteDance
Max Resolution2048px
SpeedModerate
Size Options1K, 2K
Aspect Ratios7 options

Best for: Generation and editing in one model, diverse art styles.

Coming Soon

Seedream 4.5 is not yet available. It will be enabled in a future update.

Quality Levels

GPT Image models support quality levels that control output fidelity:

LevelDescription
LowFastest, lowest cost
MediumBalanced quality and cost
HighBest quality, highest cost
AutoModel chooses optimal quality (GPT Image 1.5 only)

Higher quality levels produce more detailed images but cost more Mana.

Variation Model Override

Some models (like Imagen and FLUX.2 [pro]) are generation-only and don't support image editing natively. When you use these models for base generation, VariantLab automatically assigns a compatible model for variations:

Generation ModelVariation Model
FLUX.2 [pro]FLUX.2 [pro] Edit
Imagen 4Gemini 2.5 Flash Image
Imagen 4 UltraGemini 2.5 Flash Image

You can change the variation model in project settings if you prefer a different option.

Aspect Ratios

Available aspect ratios vary by model. Most models support 10 ratios:

RatioUse Case
1:1Avatars, icons, square art
3:2Landscape photography
2:3Portrait photography
4:3Traditional landscape
3:4Traditional portrait
5:4Wide traditional
4:5Tall traditional
16:9Widescreen, banners
9:16Vertical, mobile
21:9Ultra-wide, panoramic

GPT Image models support 3 ratios (1:1, 2:3, 3:2). Imagen models support 5 ratios (1:1, 3:4, 4:3, 9:16, 16:9).

Writing Effective Prompts

Structure

A good prompt includes:

  1. Subject - What you're generating
  2. Style - Art style, medium
  3. Details - Colors, features, accessories
  4. Composition - Position, framing
  5. Background - Setting, context

Example Prompts

Character avatar:

A cute robot mascot with large expressive blue eyes,
shiny silver metallic body, small antenna on top,
flat digital art style, centered composition,
solid white background, full body visible

Game asset:

Medieval fantasy sword with glowing blue blade,
ornate golden hilt with gems, magical particles,
game asset style, centered, transparent background,
high detail, no shadows

NFT art:

Abstract geometric lion portrait, low poly style,
vibrant gradient colors purple to orange,
modern digital art, centered, dark background

Tips for Better Results

  1. Be specific - "glowing blue LED eyes" beats "blue eyes"
  2. Include style - Always mention the art style
  3. Request centering - Helps with trait detection later
  4. Solid backgrounds - Easier for background removal
  5. Mention what to avoid - "no text, no watermarks"

Generation Settings

Remove Background

Enable to automatically remove the background after generation:

  • Uses selected background removal model
  • Replaces background with transparency
  • Can be applied/removed later

Background Removal Models

ModelBest For
U2Net FastGeneral purpose, quick
ISNet GeneralDigital art, clip art
U2Net ProHigh quality general
U2Net ClothClothing items
U2Net HumanPeople, portraits
SiluetaHigh quality general
ISNet AnimeAnime, manga style

Cost Estimation

Before generating, the button shows the estimated Mana cost for your current model and settings. Costs vary by model, image size, and quality level — faster models and smaller sizes cost less, while higher-quality models and larger outputs cost more.

Regenerating Images

Click the regenerate icon to create a new image:

  • Uses current prompt and settings
  • Replaces the existing image in that slot
  • Same Mana cost as new generation

Multiple Base Images

Generate up to 5 base images per project:

  1. Click + in the thumbnail stack
  2. Generate or upload into the new slot
  3. Star your favorite as the base for the pipeline

Having multiple options lets you pick the best starting point for your collection.