Training Tips & Best Practices

Getting Great Results from Training

Follow these best practices to ensure your custom models produce high-quality outputs.

Image Quality Matters

Resolution

• Minimum: 512 x 512 pixels

• Recommended: 1024 x 1024 or higher

• Higher resolution = more detail captured

Focus & Clarity

• Sharp, well-focused images

• Avoid motion blur

• Good exposure (not too dark or bright)

File Format

• JPG or PNG recommended

• Avoid heavily compressed files

• No watermarks or overlays

The Right Number of Images

Minimum: 10 images Optimal: 15-20 images Maximum effective: 30 images

More isn't always better—quality and variety matter more than quantity.

Variety in Training Sets

For Person Models:

• Different angles: front, side, 3/4

• Various expressions

• Indoor and outdoor

• Different outfits

For Style Models:

• Different subjects

• Same consistent style

• Various compositions

For Object Models:

• Multiple angles

• Different scales

• Various lighting setups

For Scene Models:

• Wide and close views

• Different perspectives

• Various times of day

Common Mistakes to Avoid

1. Using too few images: 10 is the minimum, aim for 15-20

2. Low variety: Same pose/angle in every image

3. Inconsistent subjects: Mixing different people/objects

4. Poor quality: Blurry, dark, or compressed images

5. Wrong model type: Choosing style when you mean object

6. Busy backgrounds: Distracting elements in training images

Checking Training Results

After training:

1. Generate test images with different prompts

2. Try various styles and settings

3. Check for consistency in outputs

4. If unsatisfied, consider retraining with better images

When to Retrain

• Results don't capture the subject accurately

• Output is inconsistent

• Model doesn't respond well to prompts

• You have better training images available