AdwolfCzar/

proxima_cena_grounded_original_dataset

Paired next-scene dataset for reference-conditioned image generation training.

Four sources, each with the same structure: dsN_/ images_A/ # control (reference) image images_B/ # target image + .txt caption

Pairs are matched by identical file stem across images_A/images_B.

subset pairs note

ds1_recortados 2,900 random subset (seed 42)

ds2_poxima_v2 5,400 main subset (90% of source… See the full description on the dataset page: https://huggingface.co/datasets/AdwolfCzar/proxima_cena_grounded_original_dataset.