ComfyUI Node: FL Kohya Dataset Config

FL Kohya Dataset Config:

The FL_KohyaSSDatasetConfig node is designed to streamline the configuration of datasets for training AI models, particularly in the context of image-based datasets. This node allows you to specify various parameters related to your dataset, such as image files, captions, and resolution settings, ensuring that your dataset is prepared correctly for training. By providing a structured way to input these parameters, the node helps in organizing and managing your dataset efficiently, which is crucial for achieving high-quality training results. The node also supports optional conditioning images and various configuration settings that can be tailored to your specific needs, making it a versatile tool for AI artists looking to optimize their training workflows.

FL Kohya Dataset Config Input Parameters:

workspace_config

This parameter specifies the workspace configuration required for the dataset. It is essential for setting up the environment where the dataset will be processed.

images

This parameter accepts image files that will be used in the dataset. The images are a crucial part of the training data and should be of high quality to ensure effective training.

captions

This parameter takes a string input for captions associated with the images. Captions provide context and additional information about the images, which can be useful for training models that require textual descriptions.

enable_bucket

This parameter allows you to enable or disable the bucketing feature. The default value is "enable". Bucketing helps in organizing images into different categories based on certain criteria, which can improve the efficiency of the training process.

resolution

This parameter sets the resolution for the images in the dataset. The default value is 1024. Higher resolutions can provide more detailed images but may require more computational resources.

num_repeats

This parameter specifies the number of times each image should be repeated in the dataset. The default value is 1, with a minimum value of 1. Repeating images can help in balancing the dataset and improving training results.

caption_extension

This parameter allows you to specify the file extension for caption files. The default value is ".caption". You can choose between ".caption" and ".txt" based on your preference.

batch_size

This parameter sets the batch size for processing the images. The default value is 1, with a minimum value of 1. A larger batch size can speed up the training process but may require more memory.

force_clear

This parameter allows you to enable or disable the force clear feature. The default value is "disable". Enabling this feature will clear existing data before processing new data, ensuring a clean workspace.

force_clear_only_images

This parameter allows you to enable or disable the force clear only images feature. The default value is "disable". Enabling this feature will clear only the images, leaving other data intact.

image_format

This parameter specifies the format of the images in the dataset. The default value is "jpg". You can choose between "png", "jpg", and "webp" based on your requirements.

dataset_config_extension

This parameter allows you to specify the file extension for the dataset configuration file. The default value is ".json". You can choose between ".toml" and ".json" based on your preference.

conditioning_images

This optional parameter accepts additional images that can be used for conditioning the dataset. These images can provide extra context or information that can be useful for training.

FL Kohya Dataset Config Output Parameters:

workspace_images_dir

This output parameter provides the directory path where the processed images are stored. This path is essential for accessing the images during the training process and ensures that all images are organized in a single location.

FL Kohya Dataset Config Usage Tips:

• Ensure that your images and captions are of high quality and accurately labeled to achieve the best training results.
• Use the enable_bucket feature to organize your images into categories, which can help in managing large datasets more efficiently.
• Adjust the resolution parameter based on your computational resources. Higher resolutions provide more detail but require more processing power.
• Utilize the num_repeats parameter to balance your dataset, especially if you have a limited number of images.
• Choose the appropriate image_format and dataset_config_extension based on your workflow and compatibility with other tools.

FL Kohya Dataset Config Common Errors and Solutions:

"Invalid image format"

• Explanation: The image format specified is not supported.
• Solution: Ensure that the image_format parameter is set to one of the supported formats: "png", "jpg", or "webp".

"Resolution out of range"

• Explanation: The resolution specified is either too low or too high.
• Solution: Set the resolution parameter to a value that is within the acceptable range, typically between 256 and 4096.

"Batch size too large"

• Explanation: The batch size specified exceeds the available memory.
• Solution: Reduce the batch_size parameter to a value that your system can handle, or upgrade your hardware to support larger batch sizes.

"Missing captions"

• Explanation: Captions are required but not provided.
• Solution: Ensure that the captions parameter is filled with appropriate text descriptions for each image.

"Invalid dataset configuration extension"

• Explanation: The dataset configuration file extension is not supported.
• Solution: Set the dataset_config_extension parameter to either ".toml" or ".json".