linoyts/animeshooter-multishot-iclora

AnimeShooter → LTX-2.3 IC-LoRA multi-shot training pairs

276 reference-guided multi-shot training pairs derived from the AnimeShooter dataset (Qiu et al., 2025, arXiv:2506.03126), formatted for the ltx-community/ltx2-lora-trainer Space in IC-LoRA mode.

Each sample is:

• videos/<id>.mp4 — a ~5.2s clip (126 frames, H.264) spanning 2–3 consecutive shots (real cuts inside the clip) from an animated YouTube video listed in AnimeShooter.
• videos/<id>_reference.png — a 768×512 reference-sheet collage: a setting panel (a frame from the same scene outside the target shots) plus up to 4 character panels (RLE-mask crops from AnimeShooter's reference-image annotations, composited on white, up to 2 views per character).
• caption — structured as ### Reference Sheet Description (per-panel positions, scene environment, character appearance from the annotations) followed by ### Target Description (chronological multi-shot description joined from the per-shot descriptive captions), matching the caption format of Lightricks/LTX-2.3-22b-IC-LoRA-Ingredients.

dataset.json is in trainer format: media_path + caption + reference_video.

Intended use: continued fine-tuning of the LTX-2.3 ingredients IC-LoRA toward multi-shot, character-consistent generation.

License & attribution

The AnimeShooter annotations are released under CC BY-NC 4.0 (academic use only); this derived dataset inherits that license. Source videos were obtained from YouTube per the AnimeShooter video-id list; copyright remains with the original owners. If any clip infringes your rights, open a discussion and it will be removed.

@misc{qiu2025animeshooter,
    title  = {AnimeShooter: A Multi-Shot Animation Dataset for Reference-Guided Video Generation},
    author = {Qiu, Lu and Li, Yizhuo and Ge, Yuying and Ge, Yixiao and Shan, Ying and Liu, Xihui},
    year   = {2025},
    url    = {https://arxiv.org/abs/2506.03126}
}