Caption file extensions (comma separated)

Leave blank to use txt, json and jsonl. Generated metadata files are excluded.

Generate training images

Generate one image per caption at each resolution before training