A Guide to LORA Training Data Collection

Without frustration or time consuming tasks :)

Using Snabbitt for Efficient Dataset Creation

Introduction

This guide outlines an optimized workflow for collecting training data for LORA models using the Snabbitt Chrome extension created by yours truly. The traditional method of manually downloading images and copying tags is time-consuming and error-prone also full of frustration. I took it upon myself to create a little extension that (for the price of a coffee) super easily allows for you to get your dataset for LORA creation within 1-5 minutes and begin training.

This guide demonstrates how Snabbitt significantly improves the training set gathering process:

Traditional Workflow (Without Snabbitt)

1. Right-click and save each image individually

2. Manually copy tags from each image page

3. Create separate text files for each image's tags

4. Ensure filenames match between images and tag files

5. Organize files into appropriate directories

6. Verify all tags were copied correctly

Estimated time per image: 1-2 minutes

Common issues: Mismatched filenames, copying errors, lost tags

Optimized Workflow (With Snabbitt)

1. Navigate to desired image on supported booru site (r34, gelbooru, e621, danbooru, etc)

2. Click Snabbitt icon > Download Image & tags

3. Image and tags are automatically downloaded and paired with the same file name.

Estimated time per image: 2-3 seconds

Benefits: Automatic filename matching, guaranteed tag accuracy

Key Advantages

1. Time Efficiency

- Reduces collection time by approximately 95%

- Eliminates manual file organization

- No need to switch between windows or applications

2. Error Reduction

- Eliminates manual copying errors

- Ensures perfect tag preservation

- Maintains consistent file naming

3. Dataset Quality

- Guarantees tag accuracy for training

- Maintains original image quality

- Preserves complete tag sets

4. Workflow Improvement

- Single-click operation

- Automatic file organization

- Immediate availability for training

Practical Example

Traditional method for 100 images:

- Estimated time: 100-200 minutes

- Multiple potential points of error

- Significant organizational overhead

Snabbitt method for 100 images:

- Estimated time: 3-5 minutes

- Zero manual error potential

- Automatic organization

Conclusion

Snabbitt transforms the LORA training data collection process from a time-intensive, error-prone task into a streamlined, efficient operation. This allows creators to focus on model training and refinement rather than dataset preparation.

Best Practices

1. Create a dedicated folder for each LORA project

2. Review downloaded content periodically

3. Respect website rate limits

4. Back up your collected datasets regularly