### ControlNet Command Line Example

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/gen_img_README.md

Example of how to use ControlNet with various parameters during image generation. Ensure ControlNet 1.0 is installed and Canny is used for preprocessing if not specified otherwise.

```batchfile
python gen_img.py --ckpt model_ckpt --scale 8 --steps 48 --outdir txt2img --xformers \
    --W 512 --H 768 --bf16 --sampler k_euler_a \
    --control_net_models diff_control_sd15_canny.safetensors --control_net_weights 1.0 \
    --guide_image_path guide.png --control_net_ratios 1.0 --interactive
```

--------------------------------

### Windows Installation Steps

Source: https://github.com/kohya-ss/sd-scripts/blob/main/README.md

Commands to clone the repository, set up a virtual environment, activate it, and install core dependencies including PyTorch.

```powershell
git clone https://github.com/kohya-ss/sd-scripts.git
cd sd-scripts

python -m venv venv
.\venv\Scripts\activate

pip install torch==2.6.0 torchvision==0.21.0 --index-url https://download.pytorch.org/whl/cu124
pip install --upgrade -r requirements.txt

accelerate config
```

--------------------------------

### Install PyTorch for CUDA 12.1

Source: https://github.com/kohya-ss/sd-scripts/blob/main/README.md

Example command to install a specific version of PyTorch for CUDA 12.1. Adjust the URL for different CUDA versions.

```bash
pip install torch==2.6.0 torchvision==0.21.0 --index-url https://download.pytorch.org/whl/cu121
```

--------------------------------

### Install pytest

Source: https://github.com/kohya-ss/sd-scripts/blob/main/tests/README.md

Install the pytest framework using pip. This is a prerequisite for running tests.

```bash
pip install pytest
```

--------------------------------

### Install xformers

Source: https://github.com/kohya-ss/sd-scripts/blob/main/README.md

Command to install the xformers library for potential performance improvements. Ensure your virtual environment is activated.

```bash
pip install xformers --index-url https://download.pytorch.org/whl/cu124
```

--------------------------------

### Example Prompt with Options

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/gen_img_README.md

Demonstrates the usage of various prompt options such as negative prompts, image dimensions, steps, and random seeds within a single generation command.

```bash
(masterpiece, best quality), 1girl, in shirt and plated skirt, standing at street under cherry blossoms, upper body, [from below], kind smile, looking at another, [goodembed] --n realistic, real life, (negprompt), (lowres:1.1), (worst quality:1.2), (low quality:1.1), bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, normal quality, jpeg artifacts, signature, watermark, username, blurry --w 960 --h 640 --s 28 --d 1
```

--------------------------------

### SD3 Network Training Command Example

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/sd3_train_network.md

Execute sd3_train_network.py using accelerate launch. Ensure all required model paths and dataset configurations are provided. Use line breaks with \ or ^ for readability.

```bash
accelerate launch --num_cpu_threads_per_process 1 sd3_train_network.py \
  --pretrained_model_name_or_path="<path to SD3 model>" \
  --clip_l="<path to CLIP-L model>" \
  --clip_g="<path to CLIP-G model>" \
  --t5xxl="<path to T5-XXL model>" \
  --dataset_config="my_sd3_dataset_config.toml" \
  --output_dir="<output directory for training results>" \
  --output_name="my_sd3_lora" \
  --save_model_as=safetensors \
  --network_module=networks.lora \
  --network_dim=16 \
  --network_alpha=1 \
  --learning_rate=1e-4 \
  --optimizer_type="AdamW8bit" \
  --lr_scheduler="constant" \
  --sdpa \
  --max_train_epochs=10 \
  --save_every_n_epochs=1 \
  --mixed_precision="fp16" \
  --gradient_checkpointing \
  --weighting_scheme="uniform" \
  --blocks_to_swap=32
```

--------------------------------

### Install ONNX Runtime

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/wd14_tagger_README-en.md

Install ONNX and ONNX Runtime for GPU acceleration. Refer to the official ONNX Runtime documentation for more installation details.

```powershell
pip install onnx onnxruntime-gpu
```

--------------------------------

### Accelerate Config Answers

Source: https://github.com/kohya-ss/sd-scripts/blob/main/README.md

Recommended answers for the 'accelerate config' prompts for a typical single-machine setup using fp16 mixed precision.

```text
- This machine
- No distributed training
- NO
- NO
- NO
- all
- fp16
```

--------------------------------

### Recommended Network Dimension Settings

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/train_leco.md

Recommended starting values for --network_dim based on model type. For SD 1.x/2.x, use 4-8. For SDXL, use 8-16.

```bash
--network_dim=4
```

```bash
--network_dim=8
```

--------------------------------

### SDXL LoRA Training Command

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/sdxl_train_network.md

Basic command line execution example for SDXL LoRA training using accelerate launch. Ensure to replace placeholders with your specific paths and configurations.

```bash
accelerate launch --num_cpu_threads_per_process 1 sdxl_train_network.py 
 --pretrained_model_name_or_path="<SDXL base model path>" 
 --dataset_config="my_sdxl_dataset_config.toml" 
 --output_dir="<output directory for training results>" 
 --output_name="my_sdxl_lora" 
 --save_model_as=safetensors 
 --network_module=networks.lora 
 --network_dim=32 
 --network_alpha=16 
 --learning_rate=1e-4 
 --unet_lr=1e-4 
 --text_encoder_lr1=1e-5 
 --text_encoder_lr2=1e-5 
 --optimizer_type="AdamW8bit" 
 --lr_scheduler="constant" 
 --max_train_epochs=10 
 --save_every_n_epochs=1 
 --mixed_precision="bf16" 
 --gradient_checkpointing 
 --cache_text_encoder_outputs 
 --cache_latents
```

--------------------------------

### Example WD14 Tagger Usage

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/wd14_tagger_README-en.md

Example of running the WD14 Tagger with a specific model repository, batch size, and training data folder.

```bash
python tag_images_by_wd14_tagger.py --onnx --repo_id SmilingWolf/wd-swinv2-tagger-v3 --batch_size 4 ..\train_data
```

--------------------------------

### TOML Configuration for Multiple Datasets

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/config_README-en.md

Example TOML configuration demonstrating how to set up multiple datasets, including DreamBooth and fine-tuning styles, with specific subset configurations for image directories, class tokens, and metadata.

```toml
[general]
shuffle_caption = true
caption_extension = '.txt'
keep_tokens = 1

# This is a DreamBooth-style dataset
[[datasets]]
resolution = 512
batch_size = 4
keep_tokens = 2

  [[datasets.subsets]]
  image_dir = 'C:\\hoge'
  class_tokens = 'hoge girl'
  # This subset uses keep_tokens = 2 (the value of the parent datasets)

  [[datasets.subsets]]
  image_dir = 'C:\\fuga'
  class_tokens = 'fuga boy'
  keep_tokens = 3

  [[datasets.subsets]]
  is_reg = true
  image_dir = 'C:\\reg'
  class_tokens = 'human'
  keep_tokens = 1

# This is a fine-tuning dataset
[[datasets]]
resolution = [768, 768]
batch_size = 2

  [[datasets.subsets]]
  image_dir = 'C:\\piyo'
  metadata_file = 'C:\\piyo\\piyo_md.json'
  # This subset uses keep_tokens = 1 (the value of [general])

```

--------------------------------

### Install DeepSpeed (Experimental)

Source: https://github.com/kohya-ss/sd-scripts/blob/main/README.md

Command to install the DeepSpeed library, which is experimental and only available on Linux or WSL2. Ensure your virtual environment is activated.

```bash
pip install deepspeed==0.16.7
```

--------------------------------

### Multi-line Captions Example (Text File)

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/config_README-en.md

Demonstrates how multi-line captions are represented in a plain text file when wildcard notation is enabled. Each line is treated as a potential caption.

```txt
1girl, hatsune miku, vocaloid, upper body, looking at viewer, microphone, stage
a girl with a microphone standing on a stage
detailed digital art of a girl with a microphone on a stage
```

--------------------------------

### Run Anima Train ControlNet LLLite

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/anima_train_control_net_lllite.md

Use this command to launch the training process. Replace placeholders with your specific model paths, dataset configuration, and output directories. Ensure you have 'accelerate' installed and configured.

```bash
accelerate launch --num_cpu_threads_per_process 1 anima_train_control_net_lllite.py \
  --pretrained_model_name_or_path="<path to Anima DiT model>" \
  --qwen3="<path to Qwen3-0.6B model or directory>" \
  --vae="<path to Qwen-Image VAE model>" \
  --dataset_config="my_anima_lllite_dataset.toml" \
  --output_dir="<output directory>" \
  --output_name="my_anima_lllite" \
  --save_model_as=safetensors \
  --cond_emb_dim=32 \
  --lllite_mlp_dim=64 \
  --lllite_cond_dim=64 \
  --lllite_target_layers=self_attn_q \
  --learning_rate=5e-5 \
  --optimizer_type="AdamW8bit" \
  --lr_scheduler="constant" \
  --timestep_sampling="shift" \
  --discrete_flow_shift=3.0 \
  --max_train_epochs=10 \
  --save_every_n_epochs=1 \
  --mixed_precision="bf16" \
  --gradient_checkpointing \
  --cache_latents \
  --cache_text_encoder_outputs \
  --vae_chunk_size=64 \
  --vae_disable_cache
```

--------------------------------

### Sample Image Generation Prompt

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/flux_train_network.md

Example prompt for generating sample images during training, including negative prompts and specific settings for guidance and CFG scale. This is used in conjunction with training commands.

```text
Japanese shrine in the summer forest. --n low quality, ugly, unfinished, out of focus, deformed, disfigure, blurry, smudged, restricted palette, flat colors --w 512 --h 512 --d 1 --l 4.0 --g 0.0 --s 20
```

--------------------------------

### Adafactor Optimizer Configuration for SDXL

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/train_SDXL-en.md

Example TOML configuration for Adafactor optimizer with a fixed learning rate, suitable for SDXL training. This setup aims to reduce GPU memory usage.

```toml
optimizer_type = "adafactor"
optimizer_args = [ "scale_parameter=False", "relative_step=False", "warmup_init=False" ]
lr_scheduler = "constant_with_warmup"
lr_warmup_steps = 100
learning_rate = 4e-7 # SDXL original learning rate
```

--------------------------------

### Loading Pre-trained LoRA Weights

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/train_network_advanced.md

Start training by loading existing LoRA weights. This is useful for fine-tuning or resuming training from a specific LoRA checkpoint. It only loads the LoRA module weights, unlike the --resume option.

```shell
--network_weights="<weight file>"
```

--------------------------------

### Textual Inversion Dataset Configuration

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/train_textual_inversion.md

Example TOML configuration file for setting up a dataset for Textual Inversion training. Specifies general settings and subset configurations including image directory, caption extension, and repeat counts.

```toml
[general]
shuffle_caption = false
caption_extension = ".txt"
keep_tokens = 1

[[datasets]]
resolution = 512                    # 1024 for SDXL
batch_size = 4                      # Can use larger values than LoRA training
enable_bucket = true

  [[datasets.subsets]]
  image_dir = "path/to/images"
caption_extension = ".txt"
num_repeats = 10
```

--------------------------------

### ControlNet-LLLite Training Configuration

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/train_lllite_README.md

Example TOML configuration for training ControlNet-LLLite models. Key parameters include model paths, training epochs, optimizer settings, learning rate, and dimensions for network and conditioning embeddings.

```toml
pretrained_model_name_or_path = "/path/to/model_trained_on.safetensors"
max_train_epochs = 12
max_data_loader_n_workers = 4
persistent_data_loader_workers = true
seed = 42
gradient_checkpointing = true
mixed_precision = "bf16"
save_precision = "bf16"
full_bf16 = true
optimizer_type = "adamw8bit"
learning_rate = 2e-4
xformers = true
output_dir = "/path/to/output/dir"
output_name = "output_name"
save_every_n_epochs = 1
save_model_as = "safetensors"
vae_batch_size = 4
cache_latents = true
cache_latents_to_disk = true
cache_text_encoder_outputs = true
cache_text_encoder_outputs_to_disk = true
network_dim = 64
cond_emb_dim = 32
dataset_config = "/path/to/dataset.toml"
```

--------------------------------

### Prompt File Format for Sample Image Generation

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/train_network_advanced.md

This format is used for specifying prompts and their associated generation parameters in a text file for sample image generation. Lines starting with '#' are comments. Options like '--n', '--w', '--h', '--d', '--l', '--g', and '--s' can be appended to each prompt.

```text
# prompt 1
masterpiece, best quality, (1girl), in white shirts, upper body, looking at viewer, simple background --n low quality, worst quality, bad anatomy,bad composition, poor, low effort --w 768 --h 768 --d 1 --l 7.5 --s 28

# prompt 2
masterpiece, best quality, 1boy, in business suit, standing at street, looking back --n (low quality, worst quality), bad anatomy,bad composition, poor, low effort --w 576 --h 832 --d 2 --l 5.5 --s 40
```

--------------------------------

### Sample pytest test function

Source: https://github.com/kohya-ss/sd-scripts/blob/main/tests/README.md

A basic example of a test function written for pytest. Test functions should start with 'test_'. This example asserts that 1 equals 2, which will fail.

```python
def test_x():
    assert 1 == 2, "Invalid test response"
```

--------------------------------

### Directory Structure Mapping to Configuration

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/config_README-en.md

Illustrates how image directories map to different configuration registration locations: [[datasets.subsets]], [[datasets]], and [general]. This visual aid helps understand the hierarchy and precedence of settings.

```text
C:\
├─ hoge  ->  [[datasets.subsets]] No.1  ┐                        ┐
├─ fuga  ->  [[datasets.subsets]] No.2  |->  [[datasets]] No.1   |
├─ reg   ->  [[datasets.subsets]] No.3  ┘                        |->  [general]
└─ piyo  ->  [[datasets.subsets]] No.4  -->  [[datasets]] No.2   ┘
```

--------------------------------

### Run pytest tests

Source: https://github.com/kohya-ss/sd-scripts/blob/main/tests/README.md

Execute all tests discovered by pytest in the current directory. Ensure pytest is installed before running.

```bash
pytest
```

--------------------------------

### Select Logging Tool

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/train_leco.md

The --log_with argument specifies the logging tool to use. Options include 'tensorboard', 'wandb', or 'all'.

```bash
--log_with="tensorboard"
```

--------------------------------

### Specify Additional LoRA Network Arguments (Conv2d with default alpha)

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/train_network.md

This example demonstrates specifying the Conv2d dimension while omitting alpha, which will default to 1.

```bash
--network_args "conv_dim=4"
```

--------------------------------

### Import AI Agent Prompts

Source: https://github.com/kohya-ss/sd-scripts/blob/main/README.md

Import the repository's recommended prompts for AI agents like Claude and Gemini by adding this line to your configuration file.

```markdown
@./.ai/claude.prompt.md
```

```markdown
@./.ai/gemini.prompt.md
```

--------------------------------

### Specify Additional LoRA Network Arguments (Conv2d)

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/train_network.md

Used for LoRA-specific parameters. This example shows how to set the rank and alpha for Conv2d (3x3) LoRA (LoRA-C3Lier).

```bash
--network_args "conv_dim=4" "conv_alpha=1"
```

--------------------------------

### Include Final Layer in LoRA Training

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/anima_train_network.md

Example of using network_args to include the final layer in LoRA training. Patterns are matched against the full module name.

```bash
--network_args "include_patterns=['.*final_layer.*']"
```

--------------------------------

### Set Attention Implementation Mode

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/anima_train_network.md

Choose the attention implementation: 'torch' (default), 'xformers', 'flash', or 'sageattn'. 'xformers' requires '--split_attn'. 'sageattn' is for inference only.

```bash
--attn_mode
```

--------------------------------

### Upgrade Repository and Dependencies

Source: https://github.com/kohya-ss/sd-scripts/blob/main/README.md

Commands to pull the latest changes from the repository, reactivate the virtual environment, and upgrade all dependencies.

```powershell
cd sd-scripts
git pull
.\venv\Scripts\activate
pip install --use-pep517 --upgrade -r requirements.txt
```

--------------------------------

### Captioning with Token String

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/train_textual_inversion.md

Example of how to format captions for Textual Inversion, ensuring the specified token string is included. This is crucial for the model to recognize the custom token.

```text
mychar, 1girl, blonde hair, blue eyes
```

--------------------------------

### Multi-Resolution Dataset Configuration

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/flux_train_network.md

Define multiple resolutions with different batch sizes in the dataset configuration file. This TOML example shows settings for two different resolutions.

```toml
[general]
# Common settings
flip_aug = true
color_aug = false
keep_tokens_separator= "|||"
shuffle_caption = false
caption_extension = ".txt"

[[datasets]]
# First resolution settings
batch_size = 2
enable_bucket = true
resolution = [1024, 1024]

  [[datasets.subsets]]
  image_dir = "path/to/image/directory"
  num_repeats = 1

[[datasets]]
# Second resolution settings
batch_size = 3
enable_bucket = true
resolution = [768, 768]

  [[datasets.subsets]]
  image_dir = "path/to/image/directory"
  num_repeats = 1

```

--------------------------------

### Recommended Settings for Chroma Models

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/flux_train_network.md

Use these settings for Chroma models, which require specific guidance scale and timestep sampling configurations.

```bash
--timestep_sampling sigmoid --model_prediction_type raw --guidance_scale 0.0
```

--------------------------------

### Dynamic Prompt: Enumeration

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/gen_img_README.md

Shows how to use all items from a list in order (enumeration) and how multiple enumerated lists generate all combinations.

```text
{e$$red|blue} flower, {e$$1girl|2girls}
```

--------------------------------

### Exclude MLP Layers from LoRA Training

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/anima_train_network.md

Example of using network_args to additionally exclude MLP layers from LoRA training. Patterns are matched against the full module name.

```bash
--network_args "exclude_patterns=['.*mlp.*']"
```

--------------------------------

### Multi-line Captions Example (JSON Metadata)

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/config_README-en.md

Shows how multi-line captions are structured within a JSON metadata file. Line breaks are represented by '\n', and tags are appended to each caption line.

```json
{
    "/path/to/image.png": {
        "caption": "a cartoon of a frog with the word frog on it\ntest multiline caption1\ntest multiline caption2",
        "tags": "open mouth, simple background, standing, no humans, animal, black background, frog, animal costume, animal focus"
    },
    ...
}
```

--------------------------------

### Template Usage for Styles

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/train_textual_inversion.md

Command-line arguments for specifying training templates. Use '--use_object_template' for characters/objects and '--use_style_template' for artistic styles.

```bash
--use_object_template
```

```bash
--use_style_template
```

--------------------------------

### Run LoRA Training Command

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/validation.md

Launch the LoRA training process using accelerate and specify all necessary parameters including model paths, dataset configuration, output directories, and training hyperparameters. Ensure logging is enabled to track validation loss.

```bash
accelerate launch sdxl_train_network.py \
  --pretrained_model_name_or_path="sd_xl_base_1.0.safetensors" \
  --dataset_config="dataset_config.toml" \
  --output_dir="output" \
  --output_name="my_lora" \
  --network_module=networks.lora \
  --network_dim=32 \
  --network_alpha=16 \
  --save_every_n_epochs=1 \
  --learning_rate=1e-4 \
  --optimizer_type="AdamW8bit" \
  --mixed_precision="bf16" \
  --logging_dir=logs
```

--------------------------------

### SDXL Training Command

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/train_leco.md

Basic command to launch LoRA training for SDXL models using accelerate. Ensure all arguments are on a single line when executing.

```bash
accelerate launch --mixed_precision bf16 sdxl_train_leco.py
  --pretrained_model_name_or_path="sdxl_model.safetensors"
  --prompts_file="slider.toml"
  --output_dir="output"
  --output_name="my_sdxl_slider"
  --network_dim=8
  --network_alpha=4
  --learning_rate=1e-4
  --optimizer_type="AdamW8bit"
  --max_train_steps=1000
  --max_denoising_steps=40
  --mixed_precision=bf16
  --sdpa
  --gradient_checkpointing
  --save_every_n_steps=200
```

--------------------------------

### Configuring Additional LoRA Network Arguments

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/train_network.md

The --network_args parameter allows specifying additional parameters specific to LoRA modules. For example, to use LoRA-C3Lier (Conv2d 3x3), you can specify 'conv_dim' and 'conv_alpha'.

```bash
--network_args "conv_dim=4" "conv_alpha=1"
```

```bash
--network_args "conv_dim=4"
```

--------------------------------

### Run FLUX.1 Training Command

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/flux_train_network.md

This command initiates the FLUX.1 model training process. Ensure all model paths and configuration files are correctly specified.

```bash
accelerate launch --num_cpu_threads_per_process 1 flux_train_network.py \
  --pretrained_model_name_or_path="<path to FLUX.1 model>" \
  --clip_l="<path to CLIP-L model>" \
  --t5xxl="<path to T5-XXL model>" \
  --ae="<path to AE model>" \
  --dataset_config="my_flux_dataset_config.toml" \
  --output_dir="<output directory>" \
  --output_name="my_flux_lora" \
  --save_model_as=safetensors \
  --network_module=networks.lora_flux \
  --network_dim=16 \
  --network_alpha=1 \
  --learning_rate=1e-4 \
  --optimizer_type="AdamW8bit" \
  --lr_scheduler="constant" \
  --sdpa \
  --max_train_epochs=10 \
  --save_every_n_epochs=1 \
  --mixed_precision="fp16" \
  --gradient_checkpointing \
  --guidance_scale=1.0 \
  --timestep_sampling="flux_shift" \
  --model_prediction_type="raw" \
  --blocks_to_swap=18 \
  --cache_text_encoder_outputs \
  --cache_latents
```

--------------------------------

### Basic Fine-tuning Command Structure

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/fine_tune.md

This is the common command structure for fine-tuning scripts across different architectures. Ensure you replace placeholders like `<path_to_model>` with your specific paths.

```bash
accelerate launch --mixed_precision bf16 {script_name}.py \
  --pretrained_model_name_or_path <path_to_model> \
  --dataset_config <path_to_config.toml> \
  --output_dir <output_directory> \
  --output_name <model_output_name> \
  --save_model_as safetensors \
  --max_train_steps 10000 \
  --learning_rate 1e-5 \
  --optimizer_type AdamW8bit
```

--------------------------------

### Converting ai-toolkit YAML to TOML

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/train_leco.md

Demonstrates the conversion of an ai-toolkit style YAML configuration to TOML format, highlighting key syntax differences.

```yaml
targets:
  - target_class: ""
    positive: "high detail"
    negative: "low detail"
    multiplier: 1.0
guidance_scale: 1.0
resolution: 512
```

```toml
guidance_scale = 1.0
resolution = 512

[[targets]]
target_class = ""
positive = "high detail"
negative = "low detail"
multiplier = 1.0
```

--------------------------------

### SD 1.x/2.x Training Command

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/train_leco.md

Basic command to launch LoRA training for SD 1.x/2.x models using accelerate. Ensure all arguments are on a single line when executing.

```bash
accelerate launch --mixed_precision bf16 train_leco.py
  --pretrained_model_name_or_path="model.safetensors"
  --prompts_file="prompts.toml"
  --output_dir="output"
  --output_name="my_leco"
  --network_dim=8
  --network_alpha=4
  --learning_rate=1e-4
  --optimizer_type="AdamW8bit"
  --max_train_steps=500
  --max_denoising_steps=40
  --mixed_precision=bf16
  --sdpa
  --gradient_checkpointing
  --save_every_n_steps=100
```

--------------------------------

### Regional LoRA Prompting Example

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/gen_img_README.md

Defines multiple prompt regions using ' AND '. The R, G, and B channels of a mask image correspond to the first three regions, respectively. Subsequent regions and negative prompts apply to the entire image.

```text
shs 2girls, looking at viewer, smile AND bsb 2girls, looking back AND 2girls --n bad quality, worst quality
```

--------------------------------

### Run Image Generation in Interactive Mode

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/gen_img_README.md

Use this command to start the image generation script in interactive mode. Specify your model checkpoint and output directory. Options like `--xformers` and `--fp16` can be used for performance and precision.

```batchfile
python gen_img.py --ckpt <model_name> --outdir <image_output_destination> --xformers --fp16 --interactive
```

--------------------------------

### Recommended Settings for FLUX.1 Training

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/flux_train_network.md

Use these settings for general FLUX.1 model training. They optimize timestep sampling and model prediction.

```bash
--timestep_sampling shift --discrete_flow_shift 3.1582 --model_prediction_type raw --guidance_scale 1.0
```

--------------------------------

### Specify Qwen-Image VAE Model Path

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/anima_train_network.md

Provide the path to the Qwen-Image VAE model.

```bash
--vae="<path to Qwen-Image VAE model>"
```

--------------------------------

### Run Lumina Network Training

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/lumina_train_network.md

Execute the lumina_train_network.py script with necessary model and dataset configurations. Ensure all paths and parameters are correctly set for your training environment.

```bash
accelerate launch --num_cpu_threads_per_process 1 lumina_train_network.py \
  --pretrained_model_name_or_path="lumina-image-2.safetensors" \
  --gemma2="gemma-2-2b.safetensors" \
  --ae="ae.safetensors" \
  --dataset_config="my_lumina_dataset_config.toml" \
  --output_dir="./output" \
  --output_name="my_lumina_lora" \
  --save_model_as=safetensors \
  --network_module=networks.lora_lumina \
  --network_dim=8 \
  --network_alpha=8 \
  --learning_rate=1e-4 \
  --optimizer_type="AdamW" \
  --lr_scheduler="constant" \
  --timestep_sampling="nextdit_shift" \
  --discrete_flow_shift=6.0 \
  --model_prediction_type="raw" \
  --system_prompt="You are an assistant designed to generate high-quality images based on user prompts." \
  --max_train_epochs=10 \
  --save_every_n_epochs=1 \
  --mixed_precision="bf16" \
  --gradient_checkpointing \
  --cache_latents \
  --cache_text_encoder_outputs
```

--------------------------------

### Configuring Multi-Resolution Datasets with Skip Image Resolution

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/config_README-en.md

Illustrates the use of `enable_bucket`, `bucket_no_upscale`, `max_bucket_reso`, and `skip_image_resolution` for managing multi-resolution training data and preventing image overlap.

```toml
[general]
enable_bucket = true
bucket_no_upscale = true
max_bucket_reso = 1536

[[datasets]]
resolution = 768
  [[datasets.subsets]]
  image_dir = 'C:\hoge'

[[datasets]]
resolution = 1024
skip_image_resolution = 768
  [[datasets.subsets]]
  image_dir = 'C:\hoge'

[[datasets]]
resolution = 1280
skip_image_resolution = 1024
  [[datasets.subsets]]
  image_dir = 'C:\hoge'
```

--------------------------------

### Basic Image Generation with Text-to-Image

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/gen_img_README.md

Generates images using a specified checkpoint, output directory, and various generation parameters. Supports xformers and fp16 for optimization. Includes detailed prompt with negative prompts.

```batchfile
python gen_img.py --ckpt trinart_characters_it4_v1_vae_merged.ckpt \
    --outdir outputs --xformers --fp16 --scale 12.5 --sampler k_euler --steps 32 \
    --image_path template.png --strength 0.8 \
    --prompt "1girl, cowboy shot, brown hair, pony tail, brown eyes, \
          sailor school uniform, outdoors \
          --n lowres, bad anatomy, bad hands, error, missing fingers, cropped, \
          worst quality, low quality, normal quality, jpeg artifacts, (blurry), \
          hair ornament, glasses" \
    --batch_size 8 --images_per_prompt 32
```

--------------------------------

### Using Negative Prompts and Prompt Weighting

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/gen_img_README.md

Demonstrates the syntax for specifying negative prompts using the `--n` flag and applying prompt weighting with parentheses and colons, similar to AUTOMATIC1111's Web UI.

```batchfile
python gen_img.py --ckpt <model_name> --outdir <image_output_destination> --xformers --fp16 --prompt "(positive prompt:1.3) --n (negative prompt)"
```

--------------------------------

### Recommended Max Train Steps Settings

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/train_leco.md

Recommended ranges for --max_train_steps. For SD 1.x/2.x, use 300-1000 steps. For SDXL, use 500-2000 steps.

```bash
--max_train_steps=300
```

```bash
--max_train_steps=500
```

--------------------------------

### Optimizer Selection and Arguments

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/train_network_advanced.md

Choose an optimizer for training, such as AdamW8bit or Adafactor. Additional arguments can be passed to the selected optimizer using key=value pairs.

```shell
--optimizer_type="AdamW8bit" --optimizer_args "weight_decay=0.01" "betas=0.9,0.999"
```

```shell
--optimizer_type="Adafactor" --optimizer_args "relative_step=True"
```

--------------------------------

### Recommended Batch Size Settings

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/train_leco.md

Recommended batch size for training. Use 1-4 for both SD 1.x/2.x and SDXL. This is typically set in the TOML configuration.

```toml
batch_size = 1
```

```toml
batch_size = 1
```

--------------------------------

### Run HunyuanImage-2.1 Training

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/hunyuan_image_train_network.md

This command launches the training process for HunyuanImage-2.1 models using accelerate. Ensure all required model paths and dataset configurations are correctly specified.

```bash
accelerate launch --num_cpu_threads_per_process 1 hunyuan_image_train_network.py \
  --pretrained_model_name_or_path="<path to HunyuanDiT model>" \
  --text_encoder="<path to Qwen2.5-VL model>" \
  --byt5="<path to byT5 model>" \
  --vae="<path to VAE model>" \
  --dataset_config="my_hunyuan_dataset_config.toml" \
  --output_dir="<output directory>" \
  --output_name="my_hunyuan_lora" \
  --save_model_as=safetensors \
  --network_module=networks.lora_hunyuan_image \
  --network_dim=16 \
  --network_alpha=1 \
  --network_train_unet_only \
  --learning_rate=1e-4 \
  --optimizer_type="AdamW8bit" \
  --lr_scheduler="constant" \
  --attn_mode="torch" \
  --split_attn \
  --max_train_epochs=10 \
  --save_every_n_epochs=1 \
  --mixed_precision="bf16" \
  --gradient_checkpointing \
  --model_prediction_type="raw" \
  --discrete_flow_shift=5.0 \
  --blocks_to_swap=18 \
  --cache_text_encoder_outputs \
  --cache_latents
```

--------------------------------

### Fine-tune Lumina-Next DiT Model

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/fine_tune.md

Command for fine-tuning Lumina-Next DiT models. It can leverage Flash Attention for faster computation.

```bash
accelerate launch --mixed_precision bf16 lumina_train.py \
  --pretrained_model_name_or_path "Lumina-Next-DiT-B.safetensors" \
  --dataset_config "dataset_config.toml" \
  --output_dir "output" \
  --output_name "lumina_finetuned" \
  --learning_rate 1e-5
```

--------------------------------

### Launch SDXL Latent Caching Script

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/train_SDXL-en.md

Launches the script to cache latents to disk for SDXL training. Ensure to specify the correct number of CPU threads per process.

```bash
accelerate launch  --num_cpu_threads_per_process 1 tools/cache_latents.py ...
```

--------------------------------

### Recommended Guidance Scale Settings

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/train_leco.md

Recommended guidance scale values for training. For SD 1.x/2.x, use 1.0-2.0. For SDXL, use 1.0-3.0. This is typically set in the TOML configuration.

```toml
guidance_scale = 1.0
```

```toml
guidance_scale = 1.0
```

--------------------------------

### Recommended Resolution Settings

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/train_leco.md

Recommended base resolution for training. Use 512 for SD 1.x/2.x and 1024 for SDXL. This is typically set in the TOML configuration.

```toml
resolution = 512
```

```toml
resolution = 1024
```

--------------------------------

### Recommended HunyuanImage-2.1 Training Settings

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/hunyuan_image_train_network.md

These are the recommended default settings for HunyuanImage-2.1 training, optimizing model prediction processing for best results.

```bash
--model_prediction_type raw --discrete_flow_shift 5.0
```

--------------------------------

### Inpainting Training Prompt with Reference Image

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/inpainting_training.md

Configure prompts for inpainting training to include a reference image for masked regions. The image path must not contain ' --'.

```text
a photo of a cat sitting on a sofa
--i /path/to/reference.jpg
--w 512 --h 512 --d 42
```

--------------------------------

### Configuring Logging Directory

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/train_network.md

Specify the --logging_dir to direct logs for tools like TensorBoard. If omitted, no logs will be generated.

```bash
--logging_dir="<log directory>"
```

--------------------------------

### Specifying Dataset Configuration File

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/train_network.md

Use --dataset_config to point to a .toml file containing dataset configurations. Using a .toml file is recommended over specifying dataset settings directly on the command line due to length.

```bash
--dataset_config="<設定ファイルのパス>"
```

--------------------------------

### Train Textual Inversion for SDXL Models

Source: https://github.com/kohya-ss/sd-scripts/blob/main/docs/train_textual_inversion.md

Launches the Textual Inversion training script for Stable Diffusion XL models. Ensure dataset configuration and model paths are correctly set.

```bash
accelerate launch --num_cpu_threads_per_process 1 sdxl_train_textual_inversion.py \
  --pretrained_model_name_or_path="path/to/sdxl_model.safetensors" \
  --dataset_config="dataset_config.toml" \
  --output_dir="output" \
  --output_name="my_sdxl_textual_inversion" \
  --save_model_as="safetensors" \
  --token_string="mychar" \
  --init_word="girl" \
  --num_vectors_per_token=4 \
  --max_train_steps=1600 \
  --learning_rate=1e-6 \
  --optimizer_type="AdamW8bit" \
  --mixed_precision="fp16" \
  --cache_latents \
  --sdpa
```