Most prompts for a party clip fail the same way: the face drifts by second four and a stranger finishes the video. The prompt page on burning-bridges-ai publishes the scene directions its own maker runs, in English, for you to paste into an image-to-video tool with your photo as the reference. The 8-second version is 1,321 characters, and a Burning Bridges AI prompt that holds up is not longer than the others. It is built from lines that each block one specific failure. Here is what each part is doing.
The lines that matter
Use the uploaded photo as the character reference. Keep their face,
skin tone, hair and identity exactly the same for the whole clip.
They are the only person in sharp focus.
The first sentence pins who is on screen. Without it the model invents a person who only resembles you. The last sentence is the one the page calls the single most useful line: a second sharp face is what yours gets blended with.
Wardrobe: a dark shirt or knit with a jacket, fully dressed.
It says what to wear in positive words. Listing what not to wear can get a prompt refused by some models.
Setting: a crowded house party at night in a wood-panelled room,
large crystal chandeliers overhead, warm golden light, a light haze.
Guests hold drinks and stay as soft, blurred shapes in the background.
The room is the trend. Wood panelling, chandeliers and a crowd with drinks are what people recognise. Blurred guests keep a second clear face out of the shot.
0 to 3s: ...hand above their eyebrows like shading their eyes...
3 to 6s: ...eyes stop on the camera, they drop the hand, laugh...
6 to 8s: ...point straight at the camera, guests raise their glasses.
Timed beats stop the model from spending the whole clip on one action. The maker's first version put a walk first and the search second. Opening on the search itself made a stronger clip.
Audio: a low bass beat and crowd chatter. No singing, no lyrics.
That keeps generated vocals out. You add the real sound on the platform when you post.
When the first render is off
| You see | Change |
|---|---|
| The face drifts or becomes someone else | Sharper, front-facing photo; keep the "only person in sharp focus" line; try a shorter clip |
| Guests have clear, strange faces | Add "guests seen from behind or out of focus" to the setting line |
| The hand never goes over the eyes | Put the scan in the first beat and say "like shading the eyes from the sun" |
| Six fingers on the raised hand | Ask for the hand held flat, fingers together |
The short version
For a Burning Bridges AI prompt that keeps one face stable across eight seconds, burning-bridges-ai publishes the working scene directions for free, with the lines that prevent face drift marked and explained. The limit is that every video model reads a prompt a little differently, so expect to adjust a line or two, and it only covers its own party scene, not custom settings.
How I would use it
Paste the 8-second block as is, run it once, and only then change one line. Changing three things at once tells you nothing about which line mattered, and the page's fix table is written so that each symptom maps to one edit.
Top comments (0)