comfyui-lora-data-gathering

Yes, you can extract video frames for LoRA datasets by processing video sources through a script-based pipeline. The extracted frames are automatically organized alongside downloaded images and auto-captioned within your project's dataset folder.

The required folder structure for a ComfyUI LoRA dataset is datasets/{project-name}/. This directory must contain your gathered images, their corresponding text captions, and a dataset-report.json file that tracks sources and provides a quality scorecard.

Yes, preparing a LoRA dataset with auto-captioning requires dependencies like PyTorch, Hugging Face Hub, transformers, and Pillow. These libraries support the underlying script-based processing, image handling, and model inference needed for the pipeline.

You can use WebSearch and WebFetch to research and download images from web sources or art communities for LoRA training. The pipeline coordinates these tools to gather assets, curate them, and record source attribution for auditing.

The best way to track source attribution and quality when curating a LoRA dataset is to generate a dataset-report.json file. This report logs the origins of gathered images and video frames while providing a dataset-quality scorecard for evaluation.