CitronLegacy/NagaDance_H3

Naga Dance — H3

Creator: CitronLegacy

Dataset Description

A tiny, hand-captioned video LoRA training set of Nagatoro (Don't Toy with Me, Miss Nagatoro) performing a short "Senpai" dance/pose routine with speech and music. Each clip is paired with a long natural-language caption that describes the outfit, pose changes, spoken lines, and audio.

Intended for:

• Nagatoro character / motion LoRA training for audio-video generation models
• Testing caption formats that include dialogue and lip-sync notes
• Small-dataset fine-tuning experiments

Dataset Structure

The dataset is uploaded as a single zip (NagaDance_H3.zip) containing one flat folder of 6 files (~6 MB): 3 videos and 3 captions.

NagaDance_H3.zip
└── NagaDance_H3/
    ├── 1.mp4
    ├── 1.txt
    ├── 2.mp4
    ├── 2.txt
    ├── 3.mp4
    └── 3.txt

Video properties

• Format: MP4, H.264 video + AAC audio
• Resolution: 768×1152 (portrait, 2:3), all clips
• Frame rate: 30 fps
• Duration: ~5.13 s each

Caption format

Each caption starts with the trigger word n4g4d4ns3, then describes outfit, pose progression, background, quoted dialogue, and music. Clips 2 and 3 also include explicit negative-style notes about audio sync (bad lip sync. lips out of sync with audio.), which were left in deliberately so the model can learn to associate them.

n4g4d4ns3, wearing a white short-sleeved collared shirt and a blue pleated skirt with white trim, smiles and shifts her pose from a slight turn to standing confidently with hands on hips, ... she says, "Gambare Gambare Senpai." ... a plain brown background, static shot. energetic music playing.

File	Content
1	Turn and hand-pose routine, music only
2	Pose change with repeated "Jaa~ Senpai." / "Senpai." lines
3	Hip-shake with "Gambare Gambare Senpai." and a final "Ho!"

Data Collection

All videos were created by CitronLegacy as fan animation of Nagatoro. The audio was not created by CitronLegacy and comes from this YouTube video. Clips were captioned by hand and with AI assistance, then re-encoded to a uniform 768×1152 / 30 fps.

Intended Uses

• Training a character/motion LoRA
• Testing dialogue- and audio-aware captioning
• Non-commercial research and education

Limitations

• Only 3 clips (~15 s total) — not enough for general training.
• One character, one outfit, plain backgrounds.
• Audio in clips 2–3 is flagged in the captions as having poor lip sync.
• Audio source: YouTube.
• Spoken lines are Japanese romanized in the captions.

Licensing

Important

The videos are original fan work by CitronLegacy. The character Nagatoro and Don't Toy with Me, Miss Nagatoro are NOT owned by CitronLegacy and belong to their respective copyright holders. The audio is NOT created by CitronLegacy; Audio source: https://www.youtube.com/watch?v=LwTmqE_yesQ. This repository is shared as a research/training-format example. Users are responsible for ensuring their use complies with applicable copyright law and the rights of the original holders.

Citation

If you use this dataset, I'd appreciate if you cite this Hugging Face repository.