The Bad Times AI prompt has to do five things at once: keep the person identical to the photo, take the moves but not the performer's face from the reference, hit six timed beats, handle paws correctly for pets, and avoid text or watermarks. On Bad Times Dance AI it is already written and tested for every dance. Make the video from one photo.
What does the Bad Times prompt have to solve?
- Identity from the photo only. The face, hair and outfit must come from your picture, and nothing from the original performer should leak in.
- Moves from the reference. The model has to copy the body movement of the reference clip without copying who is in it.
- A timeline. The routine has six beats and each lands at a set second.
- Lip-sync mouth shapes. The mouth must keep moving in a way that matches the rhythm of a vocal, even though the model does not hear the song.
- Pets are pets. Dogs and cats need paws and their own faces, not human hands.
- A clean frame. No captions, logos or watermarks drawn into the picture.
What are the six beats?
We went through the 15-second original second by second. This is the shot list a prompt has to respect.
| Seconds | Beat | What happens |
|---|---|---|
| 0.0–2.0s | Head in hands | Both hands clutch the head, eyes shut: the "bad times" opener. |
| 2.0–3.0s | Wave it off | One hand flicks out to the side, brushing the bad times away. |
| 3.0–6.0s | Hip and forehead | One hand on the hip, the other on the forehead, a slow sway with the beat. |
| 6.0–10.0s | Stop palm | An open palm pushed toward the lens, counting off on the fingers, head shaking. |
| 10.0–11.0s | The walk-in | Two quick steps straight into the camera. |
| 11.0–15.0s | Finger in the lens | Close-up: pointing at the viewer, wagging a finger and mouthing the words like an argument. |
Solo, Couple and Group are a brighter 13-second variation of this sheet. See all 6 dances.
Why do copied prompts fail?
- The face drifts. If the photo is not tied to the prompt correctly, the model blends in the performer or invents a new face.
- Hands grow on paws. Pets standing and pointing are the hardest case; a generic prompt gives a dog human fingers.
- The timing slips. Without the six-beat sheet the walk-in and the finger point arrive at the wrong second.
- No song. The video model returns picture, not the track, so a pasted prompt leaves you to add and line up the audio.
- Text appears. Models sometimes draw captions or logos unless told not to.
What is built in here?
Each dance has its own motion reference, its own prompt and its own fixed settings, tested on the Seedance 2.0 Mini, Seedance 2.0 Fast, Seedance 2.0, Seedance 2.5 models from ByteDance. We do not publish the prompt or the settings; you pick a photo and a dance, and the original audio is added after rendering. For the steps, read how to make the video. For why people ask, see the trend explained.