Musubi Dataset Config (Wan)

A ComfyUI node in musubi-tuner/wan/config with 13 inputs and 2 outputs.

By jaimitoes·Created about a year ago·Updated 12 months ago· 61

dataset_toml_path
trigger_out

◄output_filename_prefixmusubi_dataset_config►

◄general_resolution_width512►

◄general_resolution_height512►

◄general_caption_extension.txt►

◄general_batch_size1►

◄general_enable_buckettrue►

◄general_bucket_no_upscalefalse►

◄dataset1_image_directoryimage_text_path►

◄dataset1_cache_directorycache_path►

◄dataset1_num_repeats1►

◄dataset1_override_resolutionfalse►

◄dataset1_resolution_width512►

◄dataset1_resolution_height512►

Categorymusubi-tuner/wan/config

Inputs (13)

Name	Type	Default	Description
output_filename_prefix	STRING	musubi_dataset_config	Generates the training dataset with the specified settings, this will set the training resolution. Override is needed to swap the 512 default res.
general_resolution_width	INT	512	The general width for training images. This is the base resolution for your dataset, especially when buckets are disabled or for initial scaling.
general_resolution_height	INT	512	The general height for training images. This is the base resolution for your dataset, especially when buckets are disabled or for initial scaling.
general_caption_extension	STRING	.txt	The file extension for your caption (description) files (e.g., '.txt'). These files contain the text describing each image in your dataset.
general_batch_size	INT	1	The number of images processed simultaneously by the model during training. A higher batch size can speed up training but requires more VRAM.
general_enable_bucket	BOOLEAN	true	If enabled, images will be grouped into 'buckets' (different resolutions with similar aspect ratios) to optimize VRAM usage and prevent image distortion during training. Recommended to keep enabled.
general_bucket_no_upscale	BOOLEAN	false	If 'Enable Bucket' is active and this is true, images will only be downscaled to fit a bucket's resolution, never upscaled. This can prevent blurry images but might make small images too small.
dataset1_image_directory	STRING	image_text_path	If 'Enable Bucket' is active and this is true, images will only be downscaled to fit a bucket's resolution, never upscaled. This can prevent blurry images but might make small images too small.
dataset1_cache_directory	STRING	cache_path	The path to the directory containing the images for this specific dataset. Each image should have a corresponding caption file in the same directory.
dataset1_num_repeats	INT	1	The path to the directory where cached versions of the processed dataset (e.g., VAE latents, text embeddings) will be stored. Speeds up subsequent training runs.
dataset1_override_resolution	BOOLEAN	false	If enabled, this dataset will use its own specific resolution (`dataset1_resolution_width/height`) instead of the `general_resolution_width/height`.
dataset1_resolution_width	INT	512	The specific width for images in this dataset if 'Override Resolution' is enabled. Otherwise, the general resolution will be used.
dataset1_resolution_height	INT	512	The specific height for images in this dataset if 'Override Resolution' is enabled. Otherwise, the general resolution will be used.

Outputs (2)

Name	Type	Description
dataset_toml_path	STRING	—
trigger_out	*	—