UNLIMITED CAMERA ANGLES
Re-shoot one performance from as many viewpoints as you want with Seedance — same person, same words, same timing, same gestures. Only the camera moves. Master prompt, six camera blocks and the full workflow, copy-paste ready.
Want to skip the learning curve and learn directly from me?
This is how my team turns one AI clone take into a full multi-camera shoot without re-filming anything — I'll build your clone and your content system with you.
Apply to work with me →Instagram — @nicola.aiYou are not generating a new performance. You are asking the model to preserve the original video frame-for-frame and move only the camera.
One clean take becomes a wide shot, a top-down, an extreme close-up, a side profile. Cut them together and it looks like a real multi-camera shoot of one continuous performance — with zero extra filming.
Read the workflow first
Sections 1 to 4 are in order for a reason. Skipping the source preparation or the transcription is how you get drifting lips and a different face.
Copy the master prompt, swap one camera block
Paste your exact script into SPOKEN WORDS, drop one camera block into CAMERA. Keep the rest of the prompt stable. Change only the camera between generations.
One segment, one angle, then edit
Cut the source into short pieces, generate each angle separately, and build the multi-camera rhythm in the edit. Consistency comes from the edit, not from one big generation.
What you need
Five things on the table before you generate anything.
- Seedance 2.5 (or the current Seedance model that supports video reference).
- One clean reference video of the subject speaking or performing the action.
- One strong photo reference or character sheet to lock facial identity from unseen viewpoints.
- The exact spoken script from the reference clip.
- A clear camera-angle instruction for each generation.
For reliability, split the original clip into short segments and generate one new angle per segment. Reassemble the outputs in the edit. Trying to force many angle changes into one generation is faster, but less consistent.
The master prompt
One template. You only ever touch two brackets.
This is the prompt that does the work. The reference video carries the performance, the image carries the identity, the camera block carries the viewpoint, and the positive locks stop the model from inventing anything else.
@video1, keep everything the same. Re-shoot this exact clip from one new camera viewpoint. Every movement comes from @video1 and matches it frame for frame: speech, expressions, gestures, body motion. Nothing is re-performed; only the camera is new. Same person, same clothing, same location, same light. Use @image1 only to hold the facial identity from the new viewpoint. SPOKEN WORDS In @video1 they say: “[PASTE THE EXACT SPOKEN SCRIPT]”. The mouth matches these exact words at their exact original timing. CAMERA [PASTE ONE CAMERA BLOCK HERE] POSITIVE LOCKS The motion mirrors @video1 exactly in timing. Same person, same clothing, same location. Lip sync matches the spoken words. No music. No subtitles. No on-screen text.
Replace only the two bracketed parts: the spoken script and the camera block. Everything else stays word for word.
Camera blocks
Six ready-made camera instructions. Swap one in, generate, repeat.
Replace only the CAMERA section. Keep the rest of the master prompt stable. Six blocks, tested. Pick one per generation.
Camera low and tucked slightly behind a near foreground object at frame-left. The foreground object is soft and out of focus and fills roughly 20–25% of the frame. The subject remains sharp in the midground at a strong 3/4 angle. Preserve the subject’s original eyeline; they do not look into this new lens.
Camera directly above the subject, looking straight down at a true 90° top-down angle. Preserve the exact performance and environment geometry. Very slowly move the camera farther away while maintaining the vertical top-down axis.
Camera positioned just above the table surface, looking upward at the subject from a low angle. Let the table surface create strong foreground depth lines. Preserve the original eyeline toward the original camera position; the subject does not redirect their gaze into this lens.
Eye-level frontal camera on a smooth rail/dolly. During the first half, glide straight backward approximately 2 meters. During the second half, glide straight forward the same distance back to the starting position. One continuous smooth out-then-in move, no cut, no bounce.
Place the camera approximately 60–65° off the original frontal axis for a clean side three-quarter view. Keep the subject’s gaze aimed toward the original camera position. Do not rotate the subject to face the new camera.
Re-shoot from an extreme close-up framing from upper chest to top of head, with natural perspective and realistic facial detail. Keep the exact head motion, expressions, lip sync and eyeline from @video1. No artificial zoom during the shot.
Every block describes one physical camera position or one physical move, and every block tells the model where the eyeline stays. That is what keeps the subject from turning to face the new lens.
Reliable workflow
Six steps, in this order. This is where consistency is won or lost.
- Prepare the source — Use the cleanest possible original clip. Avoid hard cuts, heavy motion blur, overlays, subtitles and music if you want the model to preserve the performance accurately.
- Transcribe the exact words — Paste the exact dialogue into SPOKEN WORDS. Do not paraphrase. The script is a second timing/lip-sync anchor.
- Add the identity anchor — Use one high-quality image or a character sheet. Its job is identity preservation, not pose or performance replacement.
- Choose ONE camera idea — Start with one precise angle or camera move. Do not mix five competing camera instructions in the same generation.
- Generate short sections — For difficult shots, cut the source into smaller pieces. Generate each angle separately, then stitch the selected outputs together.
- Edit for continuity — Cut on natural gesture changes, pauses, hand movement or sentence beats. The result should feel like a real multi-camera shoot of one continuous performance.
What actually makes it work
Each input has one job. Know which is which and you can fix any output.
The video reference
Controls performance and timing.
The image reference
Protects identity when the face is seen from a new angle.
The camera block
Controls viewpoint, lens position and camera motion.
The positive locks
Prevent the model from inventing a new scene, outfit, performance or edit.
Shorter generations
Reduce drift and make continuity easier to control.
Troubleshooting
Six things that go wrong, and the exact line that fixes each one.
The face changes
DiagnosisThe identity anchor is too weak for the angle you asked for.
FixUse a stronger identity image/character sheet. Reduce the angle extremity. Explicitly say @image1 is used only to hold facial identity from the new viewpoint.
The subject looks into the new camera
DiagnosisThe eyeline was not locked.
FixLock the eyeline: “The subject keeps looking toward the original camera position and never redirects their gaze into this new lens.”
Gestures change
DiagnosisThe model re-performed instead of copying.
FixReinforce: “Every movement comes from @video1 and matches frame for frame. Nothing is re-performed.” Shorten the source segment.
The room changes
DiagnosisThe locks on location and background were not repeated, or the angle reveals space the model has never seen.
FixRepeat: same location, same object placement, same lighting, same background geometry. Avoid asking for a camera position that would reveal impossible/unseen space unless the model has enough reference information.
Lip sync drifts
DiagnosisThe words are not exact, or the clip is too long.
FixPaste the exact spoken words and use shorter clips. Do not add new dialogue.
The camera move is messy
DiagnosisThe camera instruction is vague.
FixDescribe one physical move with a start, direction, distance and end. Avoid vague language like “cinematic dynamic camera.”
If consistency matters more than speed: one clip segment = one angle. Generate clean pieces, then create the multi-camera rhythm in the edit.
Quick prompt formula
Five ingredients. If one is missing, the output drifts.
Reference performance carries the timing. Exact words anchor the lip sync. One camera block sets the viewpoint. The identity anchor holds the face. The positive locks keep everything else exactly as it was.
One take in, unlimited angles out. Clone once. Create without filming every day.
One clip segment = one angle.
- Reference video controls the performance
- Exact spoken words anchor the lip sync
- One camera block per generation
- Identity image only holds the face
- Positive locks keep scene, outfit and timing
REFERENCE PERFORMANCE + EXACT WORDS + ONE CAMERA BLOCK + IDENTITY ANCHOR + POSITIVE LOCKS
Want to skip the learning curve and learn directly from me?
Clone once, create without filming every day — I'll build your AI clone and the whole content system with you, so your expertise stops being invisible.
Apply to work with me →Instagram — @nicola.ai