Seedance 2.5 Dialogue Scenes: Keep Screen Direction and Eyelines Consistent
To make separate dialogue shots feel like two people are facing each other, establish their positions first, then state each person’s facing direction, eyeline target, and background anchor in every prompt. Generate the shots separately, inspect the actual frames, and only then cut them together. Maya and Leo are labels used in this example, not platform identity locks.
What should you decide before generating?
Start with a simple table-side conversation. Maya wears a green cardigan and sits on the left side of the table from the audience’s view. Leo wears a rust-colored shirt and sits on the right. Screen-left and screen-right describe the viewer’s frame; a person’s own left and right are different coordinates.
The 180-degree axis is a continuity convention that helps viewers understand where people are. It is not an absolute ban on crossing the line. Do not confuse this rule with a 180-degree shutter angle, which is a camera-exposure term. Adobe explains how continuity editing uses consistent screen direction and eyelines to support spatial clarity: https://www.adobe.com/creativecloud/video/hub/ideas/what-is-continuity-editing-in-film.html
How can three shots establish a believable exchange?
Use a wide shot to establish geography, then keep the eyeline targets explicit in each close shot. The following plan is a teaching example, not a tested result.
| Shot | Appearance and position | Eyeline and camera | Background anchor | Action | Connection condition |
|---|---|---|---|---|---|
| Wide | Maya in green cardigan, screen-left; Leo in rust shirt, screen-right | Camera faces the table from the room entrance | Tall plant behind Maya; framed print behind Leo | Both settle and turn slightly toward each other | End with both profiles and the table edge visible |
| Maya close-up | Maya fills the left half of frame | Maya looks toward screen-right, off-camera at Leo | Plant remains over her screen-left shoulder | Maya listens and gives one small nod | Cut when her gaze is steady toward Leo |
| Leo reverse | Leo fills the right half of frame | Leo looks toward screen-left, off-camera at Maya | Framed print remains over his screen-right shoulder | Leo listens, gives one brief nod, and keeps looking toward screen-left. | Begin with his eyeline matched to Maya’s actual position; equal eye height within the frame is not required. |
The subjects need not occupy identical pixel positions in every shot. Compare gaze direction, eye height, shoulder orientation, and recognizable background anchors instead.
What should each prompt contain?
Each prompt must stand on its own. Upload or reuse the relevant reference image for that shot; do not assume the generator remembers a previous clip. Keep the action small so the spatial relationship remains easy to review.
Wide establishing shot
Reference preparation: A clear two-person table scene with both figures separated and the plant and framed print visible.
Use the supplied two-person table image as the visual reference. Establish Maya in a green cardigan seated on screen-left and Leo in a rust-colored shirt seated on screen-right, viewed from the room entrance. Keep the tall plant behind Maya and the framed print behind Leo. Both people face generally toward the other across the table. Fixed medium-wide camera, natural room light, no dialogue. Allow only a small settling movement and one brief glance between them. Preserve the table edge, seating order, and screen-left/screen-right relationship.
Check: Count people, confirm the anchors remain on the same sides, and note the final gaze direction. If the seating order changes, simplify the action before making close shots.
Maya’s listening close-up
Reference preparation: A portrait of Maya with the plant visible behind her screen-left shoulder and enough space on screen-right for an off-camera eyeline.
Use the supplied Maya portrait as the visual reference. Frame Maya in a medium close-up on the screen-left side of the composition, with the plant behind her screen-left shoulder. Maya faces slightly toward screen-right and looks at Leo off-camera at eye level. Keep her green cardigan, face, shoulder angle, and soft room lighting stable. She listens silently, then gives one small nod while keeping her gaze toward screen-right. Fixed camera, no dialogue, no turn away, no new person, no background rearrangement.
Check: The nod should not redirect her gaze. If she looks at the lens, change only the eyeline sentence and test again.
Leo’s reverse listening shot
Reference preparation: A portrait of Leo with the framed print behind his screen-right shoulder and a clear screen-left eyeline.
Use the supplied Leo portrait as the visual reference. Frame Leo in a medium close-up on the screen-right side of the conversation, with the framed print behind his screen-right shoulder. Leo faces slightly toward screen-left and looks at Maya off-camera at eye level. Keep his rust-colored shirt, face, shoulder angle, print position, and soft room lighting stable. He listens silently, gives one brief nod, and keeps looking toward screen-left. Fixed camera, no dialogue, no profile reversal, no new person, no background rearrangement.
Check: Compare Leo’s eye height with Maya’s shot and confirm that the print stays on his screen-right. If the reverse feels like the same direction as Maya’s shot, revise Leo’s eyeline rather than flipping the whole frame.
How do you diagnose a broken reverse shot?
Change the smallest relevant instruction, then review the affected shot and its neighboring cut.
| Symptom | Edit which shot? | Change which instruction? | Recheck |
|---|---|---|---|
| Both people look toward the same side | The shot with the wrong gaze | State the off-camera target as screen-left or screen-right | Gaze direction against the wide shot |
| Position feels confusing after the cut | The close-up that loses geography | Repeat the correct background anchor and shoulder orientation | Anchor side and table edge |
| Gaze does not align with the other character’s position | Usually the reverse shot | Check whether each gaze fits the characters’ actual relative positions; do not require equal screen heights | Compare gaze direction with subject height, shot size, and camera position |
| Background object swaps sides | The shot with the swapped anchor | Name the anchor’s screen side explicitly | Plant/print side before and after cut |
| Appearance is right but cut still fails | Both neighboring shots | Compare facing angle and final/starting gaze, then trim unusable frames | Silent playback in sequence |
A person can move from one side of the frame in a tighter composition without breaking continuity. The useful question is whether the viewer can still infer the same room and facing relationship.
Why not simply mirror every wrong shot?
Horizontal flipping can reverse product text, asymmetric clothing details, a person’s habitual hand, and the room’s geography. Consider rebuilding the wide shot when the seating order is unclear, regenerating one close-up when its eyeline is wrong, or trimming to frames that preserve the relationship. External editing can improve a cut, but it cannot guarantee that a mirrored or stabilized image represents the intended space.
What is the safest review order?
First inspect each generated shot alone for people, faces, and background anchors. Next play the three shots in order with sound off. Check screen direction, eye height, and the action at each cut. Only after the silent sequence works should you add dialogue and other audio. Prompt goals describe intentions; they do not prove the output achieved them.
For planning a more detailed shot list, see the AI video shot list template. The multi-character prompt guide covers assigning actions within one frame, while the character consistency guide focuses on appearance across shots. Start a new test from the Seedance 2.5 workspace.
FAQ
Can one reference image guarantee a consistent shot-reverse-shot pair?
No. It can clarify appearance and setting, but each shot still needs explicit position, eyeline, and anchor instructions, followed by frame review.
Why do two people look as if they are not looking at each other?
They may share a screen direction, have eyelines that do not match each other’s actual positions, or be framed from different sides of the axis. Compare the actual gaze and background anchors rather than relying on the prompt wording.
When can you cross the axis?
A deliberate move, a neutral front-facing shot, or a clearly shown camera transition can reset the viewer’s geography. Use a wide or neutral bridge when a cut would otherwise be confusing. Match the gaze to the other character’s actual position; equal eye height within the frame is not required.
Should you generate the whole conversation at once?
Use one continuous generation only when its available controls and the resulting footage meet your review needs. Separate shots give you more control over testing each eyeline and cut, but they require careful matching and external editing.
