跳转到主要内容

AI Model Studio

Train custom AI models (LoRA) on your products, characters, or styles. Upload images and create a personalized AI model that generates content in your specific style. Availability: Agency Plan and above
Training Time: 30-60 minutes
Output: Custom LoRA model

使用方法

1

Go to AI Model Studio

2

Create New Model

Click + New Model and name your model
3

Upload Training Images

Upload 10-20 high-quality images:
  • Product photos from multiple angles
  • Consistent lighting preferred
  • Clear, focused images
  • Same subject/theme across all images
4

Configure Training

  • Set trigger word (e.g., “myproduct”)
  • Adjust training steps (default: 1000)
  • Select base model
5

Start Training

Click Train Model and wait 30-60 minutes
6

Use Your Model

Once trained, use in Image Generation with your trigger word

Training Tips

Image Quality

  • High resolution (at least 1024px)
  • Consistent lighting
  • Multiple angles and poses
  • Clean backgrounds (or consistent backgrounds)

Quantity

  • Minimum: 10 images
  • Recommended: 15-20 images
  • More is not always better (quality over quantity)

Consistency

  • Same subject across all images
  • Similar style/aesthetic
  • Consistent product type (don’t mix unrelated items)

使用场景

Product Lines

Train on your product line to generate consistent product images:
  • New product concepts
  • Variations of existing products
  • Product in different contexts

Brand Characters

Create consistent brand characters:
  • Mascots
  • Brand ambassadors
  • Consistent avatars

Style Training

Train on artistic style:
  • Brand aesthetic
  • Photography style
  • Design language

Using Your Model

After training, use your model in Image Generation:
  1. Go to Image Generation (/app)
  2. Select your custom model from the dropdown
  3. Include trigger word in prompt: photo of myproduct on table
  4. Generate images in your trained style

最佳实践

Clear Subject: Ensure all training images show the same clear subject.
Good Lighting: Consistent, well-lit images train better models.
Test and Iterate: Train, test results, add more images if needed, retrain.

故障排除

  • Poor results: Add more diverse angles and lighting
  • Overfitting: Reduce training steps or use fewer images
  • Inconsistency: Ensure training images are consistent

下一步