"Tsuhan no Oni" — 15-second sung jingle CM
A 15-second sung CM in which the mail-order mascot "Tsuhan no Oni" and four colour-variant friends dance in unison. One jingle from ElevenLabs Music, dance shots from image-to-video on colour-variant key images, and a logo end card, all joined in ffmpeg — for about ¥85 in generation cost.
| ElevenLabs Music | One sung jingle; lyrics written entirely in kana |
|---|---|
| Kling 3.0 (image-to-video) | Colour-variant key images → synchronised dance, 0:00–0:12 |
| ffmpeg | Real logo composited on a white rounded panel as the CTA end card; concatenation |
| Keyless AI | Keyless route to Higgsfield and ElevenLabs |
How it was made
The first sample to run the whole chain — jingle → character images → i2v dance → concatenation. Lyrics are written entirely in kana ("deta! tsuhan no oni!") so the singing model cannot misread them, and the CTA text is a composited real logo rather than AI-drawn type.
Model choice is on the record too: Seedance 2.0 refused the characters as detected IP, so the job was switched explicitly to Kling 3.0. The rule "never fall back silently when a model fails" comes from here.
- 15-second cut completed (≈37 Higgsfield credits + one Music generation)
- Registered as a sample on docs (first entry of jingle-cm-video)