1. Download and Install Software and Models
   1. Download “RAW” SD models
2. Preparing training datasets
   1. 20-30 images
   2. UP than 512 px X 512 px
   3. same ratio 512X512
3. tags the training datasets
   1. training labels
4. Training LoRA
5. Test LoRA

SDXL + LoRA

Fine tuning後的模型檔案與搭配使用的基礎模型

flowchart TD
    A[初步資料蒐集：收集 1,600 張圖像]
    B[情境分類：依據 160 個情境案例進行分類]
    C[專家與使用者檢視]
    D[候選圖像確認：每個情境至少選取 1-2 張]
    E[資料標籤設定]
    F[最終資料集確認與版本管理]
    G[選定預訓練模型 例如 Stable Diffusion]
    H[LoRA 微調設定 只更新關鍵層]
    I[使用最終資料集進行模型微調訓練]
    J[驗證與調整：專家、使用者反饋]
    K[最終 LoRA 模型產出]

    A --> B
    B --> C
    C --> D
    D --> E
    E --> F
    F --> G
    G --> H
    H --> I
    I --> J
    J --> K

flowchart TD
    A[Initial Data Collection] --> B[Scenario Classification]
    B --> C[Expert Review and Screen]
    C --> D[Candidate Image Selection]
    D --> E[Data Labeling]
    E --> F[Final Dataset Confirmation & Version Control]
    F --> G[Select Pre-trained Model]
    G --> H[Fine-tune the Model Using the Final Dataset]
    H --> I[Validation and Adjustment: Incorporate Expert and User Feedback]
    I --> J[Final Fine-tuned Model Output]