Cross-model image-to-video (I2V) generation pairs for training video reward
models (RM) and DPO-style preference learning.
The HF Dataset Viewer renders each row as prompt + two videos side-by-side.
All videos are generated I2V — text caption + first frame of source_video
→ video. For every pair, both models receive the same prompt, same source
frame, and same seed, so the only varying axis is the model identity itself.… See the full description on the dataset page:
https://huggingface.co/datasets/qgfvadfuvads/i2v_data_v2.
