Skip to content

notebook-workflow.md

Latest commit

 

History

History
77 lines (53 loc) · 2.91 KB

notebook-workflow.md

File metadata and controls

77 lines (53 loc) · 2.91 KB

Notebook Workflow Guide

Overview

The LoRA Easy Training system uses three specialized notebooks for different stages of the training process:

  • Dataset_Maker_Widget.ipynb - Prepare images and captions for training
  • Unified_LoRA_Trainer.ipynb - Configure and execute LoRA training
  • Utilities_Notebooks.ipynb - Calculate training parameters and resize LoRA models

Data Ingestion Options

URL/ZIP Download

Download and extract datasets from URLs (Hugging Face, Civitai) or local ZIP files through the widget interface.

Direct Image Upload

Upload individual images directly into your dataset folder using the file upload widgets.

Gallery-DL Scraper

Use the integrated gallery-dl tool to scrape images and metadata from over 300 supported websites.

Getting Model and VAE Links

To use custom models or VAEs, you need direct download links. Here's how to find them:

From Civitai

Method 1: Using Model Version ID

  1. Navigate to the model or VAE page
  2. Check the URL for ?modelVersionId=XXXXXX
  3. Copy the entire URL if the ID is present
  4. If no ID is visible, switch between model versions to make it appear

How to get a link from Civitai using the version ID

Method 2: Copying Download Link

  1. Scroll to the "Files" section on the model page
  2. Right-click the Download button
  3. Select "Copy Link Address" from the context menu

How to get a link from Civitai by copying the download address

From Hugging Face

Method 1: Repository URL

  1. Go to the model or VAE repository main page
  2. Copy the URL from your browser's address bar

How to get a link from Hugging Face using the repository URL

Method 2: Direct File Link

  1. Navigate to "Files and versions" tab
  2. Find the specific file you want
  3. Click the "..." menu next to the file
  4. Right-click "Download" and copy the link address

How to get a link from Hugging Face by copying the direct file address

Advanced Features

Image Utilities

  • Resizing: Batch resize images to target resolutions with quality options
  • Quality optimization: Adjust compression and quality settings

Tag Curation

  • FiftyOne Integration: Visual tag editing interface for dataset inspection
  • Batch operations: Apply changes to multiple images simultaneously
  • Caption management: Edit and refine training captions

System Architecture

The unified architecture features:

  • Automatic model detection: Identifies SDXL vs SD 1.5 models automatically
  • Kohya backend integration: Uses proven training strategies and scripts
  • Cross-platform compatibility: Works with conda, venv, and system Python installations
  • Memory optimization: Automatic VRAM detection and profile selection
  • Environment agnostic: Supports local, VastAI, and RunPod deployments