For the specific workflow shown in the tutorial video, use a source video as the motion and environment anchor, one image for the lead actor, and one image for the repeated crowd identity. This version is deliberately strict: it tells the model exactly which person to replace, which people to clone, and which visual elements must remain unchanged. You can reuse the structure with other scenes by changing only the role descriptions and reference labels.
CHARACTER REPLACEMENT TASK
INPUTS
Source Video: @video1
Lead Actor Reference: @image1
Crowd Clone Reference: @image2
SOURCE FOOTAGE — @video1
Use @video1 as the absolute base video for tracking, choreography, blocking, framing, camera movement, timing, pacing, and environment continuity.
Preserve the original action frame by frame. Match the source motion, speed, body positioning, spatial relationships, and timing as closely as possible. Do not reinterpret or redesign the choreography.
LEAD CHARACTER — @image1
Identify the specific male character in @video1 wearing the white shirt who:
1. begins walking forward with his back facing the camera,
2. moves through the crowd,
3. turns around,
4. stops in the middle of the crowd,
5. performs the cigarette-smoking action.
Replace ONLY this walking-then-smoking character with @image1.
Maintain the exact identity characteristics of @image1 throughout the full shot, including facial structure, hairstyle, skin tone, body build, proportions, and overall appearance. Preserve identity through rear view, side angles, the turn, the stop, and the smoking action.
Do NOT apply @image1 to any stationary person, background student, foreground student, side-row student, or dancing crowd member.
CROWD REPLACEMENT — @image2
Replace EVERY student wearing the black blazer school uniform with an identical clone of @image2.
This applies to all crowd positions, including:
- background students,
- foreground students,
- left and right side rows,
- partially visible students,
- dancing students,
- students positioned behind or around the lead character.
All of these crowd members must use the identity from @image2 while retaining their original body positions and the aggressive synchronized torso and head choreography from @video1.
Do not leave any original uniformed student unchanged.
ENVIRONMENT — LOCKED
Keep the original environment from @video1 unchanged.
Preserve exactly:
- the brick school entrance,
- the steps,
- the signage,
- the architecture,
- the spatial layout,
- the background structure,
- the lighting relationship,
- the scene perspective.
Do not redesign, replace, stylize, remove, or add environmental elements.
CAMERA — LOCKED
Retain the original continuous camera pull-back and zoom-out behavior from @video1.
Match the original camera path, framing progression, lens feel, speed, distance changes, timing, and composition as precisely as possible. Do not introduce new camera angles, cuts, pans, tilts, pushes, shakes, or reframing.
ACTION CONTINUITY
Preserve the full original performance and sequence of actions. The lead must complete the walk, turn, stop, and cigarette-smoking action at the same moments as the source. The crowd must retain the same synchronized aggressive movement and timing.
IDENTITY PRIORITY
@image1 = lead character only.
@image2 = every black-blazer student only.
Never mix these identities.
Never apply @image1 to the crowd.
Never apply @image2 to the lead.
AVOID
- applying @image1 to stationary students,
- leaving any original black-blazer crowd actor unchanged,
- identity blending between @image1 and @image2,
- face distortion or identity drift,
- changing the lead's body build,
- skipping or altering the cigarette-smoking action,
- changing choreography, speed, timing, or body placement,
- altering the brick entrance, steps, signage, architecture, or background,
- changing the continuous pull-back / zoom-out camera movement,
- adding extra people, props, actions, cuts, or environmental details.
FINAL OUTPUT
The final result should look like the exact same shot from @video1 in terms of choreography, camera, timing, environment, framing, and action — with only two intentional visual changes: the walking-and-smoking lead is replaced by @image1, and every black-blazer student is replaced by clones of @image2.
ChatGPT for prompt writing and scene breakdown