Complete Guide to AI Models
Model Architectures
Flux (v3/v3.1)
Latest architecture with enhanced image quality and generation speed. Available as:
- Flux Schnell (v3.1): Optimized for speed
- Flux Dev (v3): Highest quality, supports custom model training
SDXL (v2)
Stable Diffusion XL architecture with:
- Better composition understanding
- Superior skin texture rendering
- Support for multiple styles
SD 1.5 (v1)
Classic Stable Diffusion architecture:
- Proven reliability
- Extensive LoRA support
- Most compatible with existing resources
Feature Comparison
| Feature | v3.1 (Flux Schnell) | v3 (Flux Dev) | v2 (SDXL) | v1 (SD 1.5) |
|---|---|---|---|---|
| Base Resolution | 1024x1024 | 1024x1024 | 1024x1024 | 512x512 |
| Credits per Image | 2 | 5 | 2 | 1 |
| Public Trained Characters/Styles | ❌ | ✅ | ✅ | ✅ |
| ControlNet | ✅ | ✅ | ✅ | ✅ |
| HighRes Fix | ❌ | ❌ | ✅ (+5 credits) | ✅ (+5 credits) |
| Train Model | ❌ | ✅ | ❌ | ❌ |
Model Comparisons
Photorealistic Models
professional headshot portrait of a woman wearing a black suit, natural lighting
professional headshot portrait of a man wearing a black suit, natural lighting
3D Render Models
3d render of cute robot character, pixar style
3d render of futuristic cityscape, cyberpunk style
Anime Models
anime style portrait of a young woman with blue hair
anime style portrait of a anime couple, romantic scene
Understanding Model Differences
Each model has unique characteristics that make them better suited for different tasks. Here's what you should know:
Celebrity Recognition
How well each model recognizes and generates celebrity likenesses:
- ✓V3 & V3.1 (Flux):
Better at recognizing and recreating celebrity features and characteristics
- ⚠V2 (SDXL):
May create generic faces matching basic descriptions (ethnicity, age, etc.)
- ⚠V1 (SD 1.5):
Often creates generic faces with basic matching characteristics
Artistic Style Handling
How different models interpret artistic styles:
- ✓V3 & V3.1 (Flux):
Better overall composition with balanced artistic interpretation
- ✓V2 & V1:
More expressive and dramatic artistic interpretations
Human Features Quality
How well each model handles human anatomy:
- ✓V3 & V3.1 (Flux):
Most accurate human features with minimal issues
- ⚠V2 (SDXL):
Generally good but may struggle with hands and complex poses
- ⚠V1 (SD 1.5):
May have noticeable anatomical inconsistencies
Content Safety Features
How models handle content filtering:
- ✓V3 & V3.1 (Flux):
Strong content filtering and safety features
- ⚠V2 & V1:
May require more specific prompting to maintain desired content level
Reference Image Accuracy
How well models maintain reference image characteristics:
- ✓V3 (Flux Dev):
Best at maintaining facial features from reference images
- ✓V3.1 (Flux Schnell):
Good reference maintenance with faster processing
- ⚠V2 (SDXL):
Decent reference maintenance but may vary
- ⚠V1 (SD 1.5):
May significantly deviate from reference images
Best Practices & Tips
- Use v3.1 (Flux Schnell) for rapid prototyping and iterations
- Use v3 (Flux Dev) for final high-quality images and custom trained models
- Consider v2 (SDXL) for portraits where skin texture is crucial
- Use v1 (SD 1.5) when you need our list of extensive styles and trained characters or working with legacy resources
- Enable HighRes Fix (+5 credits) on v1/v2 models for larger outputs