Underground Rapper Video

Ai artists manifest in Seedance 2.5

Pushed the new model to its limits with this dynamic AI music video.

▾Full prompt below and few takeaways:

- Prompt adherence has improved significantly.
- Audio artifacts are almost completely gone.
- Generating a full 30-second video in one pass is still risky — control and smaller details start to fall apart.
- Omni Reference reinterprets the input audio instead of preserving it exactly. For precise synchronization, separate shorter generations are still the safer workflow.
- I couldn’t stop the model from repeating parts of the lyrics. Adding the lyrics to the prompt actually made it worse, sometimes creating duplicated lines and extra voices.
- Local editing tools probably won’t solve that particular issue.

So be careful with your credits when generating 30-second clips.

@BytePlusGlobal Still, the result is kind of staggering — especially the physical consistency, camera dynamics, character performance, and environmental motion.

Made in @capcutapp with GPT-2 for references.

Prompt

REFERENCE SCOPE

@ IMAGE1 defines only the artist’s identity and wardrobe. Preserve her face, platinum-white blunt bob, round black sunglasses, slim athletic build, long narrow black coat, oversized white T-shirt, short black technical shorts, heavy black boots, and the small white H on the front left chest.

@ IMAGE2 defines the studio: matte black floor and walls, monumental angled white acoustic structures, large white ceiling panels separated by deep cosmic gaps with restrained violet-blue nebulae, and the sculptural cube-like rear wall.

@ IMAGE3 defines the exact first frame: composition, symmetry, camera position, artist scale and darkness. The ceiling must follow @ IMAGE2.

@ VIDEO1 is the sole fixed audio reference and exact performance-timing reference. Ignore its black visual track.

AUDIO

Keep the audio from @ VIDEO1 unchanged at its original timestamps from beginning to end.

All cuts and camera changes affect only the image. They must not alter the audio timing.

Do not generate any new audio, vocals, voices, sound effects, ambience, room tone, impacts or transition sounds.

DIRECTION

Create a cinematic photorealistic underground female rap performance video, 16:9, 1080p.

Keep the artist recognizable and dominant. Her movements feel natural and confident: brief stillness, loose asymmetric gestures, weight shifts, torso turns, short purposeful steps and occasional approaches toward the lens.

No repetitive foot tapping, stomping or looping gestures.

Use lively human handheld camera work: pushes, recoils, brief impact jolts, small tilts, slight overshoots and quick corrections. Keep her face readable. Avoid continuous random shaking and smooth floating gimbal movement.

SHOT STRUCTURE

0–4 SECONDS — OPENING

Start exactly on @ IMAGE3 with no pre-roll or alternate opening frame.

She prepares in darkness with her mouth closed. The studio wakes suddenly in one synchronized lighting event.

The camera immediately makes one fast physical rush-in with natural motion blur, reaching a tight frontal medium close-up as the performance begins. No cut before this moment.

4–8 SECONDS — FRONTAL CLOSE PERFORMANCE

Tight frontal medium close-up. She performs directly toward the lens with natural varied gestures.

The camera stays close, reacting to her forward movement with short pushes and recoils. Ceiling panels clearly shift between dark, dim and bright groups.

HARD CUT.

8–12 SECONDS — LOW-ANGLE PERFORMANCE

Low wide-angle medium shot near the floor. She remains large in frame, takes one purposeful step and leans toward the lens.

Use one strong handheld jolt, a brief horizon tilt and a fast correction.

WHIP CUT motivated by her movement.

12–16 SECONDS — SIDE STRUCTURE

Three-quarter side performance beside one monumental white acoustic structure.

The camera tracks diagonally closer while she turns toward the lens. The nearby structure shifts subtly once and the ceiling lighting continues changing.

HARD CUT.

16–20 SECONDS — EXTREME CLOSE PERFORMANCE

Frontal close-up with strong proximity.

The camera pushes close, recoils, briefly moves off-axis and recovers her face. Her performance remains natural, intense and varied.

End with a clean hard cut. No palm covering or touching the lens.

20–24 SECONDS — CUBE WALL PERFORMANCE

Close angled performance in front of the exact sculptural cube rear wall from image2.

The cube relief moves in broad connected mechanical waves with visible depth, weight and changing shadows. Nothing detaches or flies.

The camera stays close and handheld while she performs in front of the moving wall.

HARD CUT.

24–30 SECONDS — FINAL CLIMAX

Close hero performance at the exact cube rear wall. Move progressively closer throughout this final shot.

The cube wall, side structures and ceiling lighting reach their strongest controlled kinetic state while the studio remains physically intact.

The camera becomes more energetic with brief jolts, forward pressure and quick corrections, while keeping her face clear and dominant.

Near the end, she smoothly removes her sunglasses and reveals beautiful realistic eyes with subtle natural violet irises. No glow, emitted light or supernatural effects.

She holds direct eye contact for the final beat while the studio continues moving softly behind her.

End in a tight close portrait. No fade, no zoom-out and no empty environment shot.

LOCKS

Follow the stated shot order and framing.

Keep the artist visible, recognizable and dominant in every shot.

Use @ sound1  only as the fixed audio and performance-timing reference.

Do not include written lyrics anywhere in the generated video.

No additional audio of any kind.

No portal, black hole, fantasy reveal, studio destruction, flying cubes or detached fragments.