AI video mirror reflection prompts: describe one person in two linked views.
Identify the person and mirror in the starting frame, describe one visible action, and ask for its reflection to occur at the same time. Keep the view simple enough to inspect both. Then check the actual clip: a prompt can express the relationship, but cannot guarantee matching identity, geometry or timing.
1. Separate what the source shows from what you need
Before writing motion, note the visible person, the mirror boundary, and the reflected view. Is the person's face visible directly, in the mirror, or both? Which clothing or hair details can you actually compare? If the opening reflection already shows a different person or an implausible pose, prepare a better reference rather than treating that mismatch as an animation instruction.
A mirror gives another view of the scene. Do not require the person and reflection to look like identical pasted images: a camera behind the person may see their back while the mirror shows their face. Plan a shot where the details that matter can be seen, and judge changes against that starting arrangement.
- Person: visible position, pose and identifying details.
- Mirror: border, surface and the area that contains the reflection.
- Reflection: visible face, clothing and starting pose.
- Action: what should move, and the pose where it should end.
- View: what stays visible when the action happens.
Make these observations in your own notes or image editor. Shotmelo does not mark mirror regions or calculate reflection geometry. For general reference preparation, see the framing and crop guide.
2. Name the reflected action without adding a second actor
Write the person's action first, then the relationship you expect to see in the mirror. For example: the person makes a small nod and returns to the opening pose; the mirror shows the same person making that nod at the same time. A direction such as “the other person in the mirror smiles” can describe a different story from the one you intend.
The MirrorWorld research preprint treats reflected content and its spatial arrangement as separate challenges in video reflection generation. That is useful context for checking both who appears and where they appear. It does not establish that a particular phrase fixes a commercial generator, or that the research method is available in your selected tool.
Runway's image-to-video prompting guide describes the input image as establishing the scene and recommends starting with simple motion. Apply that distinction here: use the reference for the opening arrangement, and the text for the action. Neither substitutes for inspecting the finished reflection.
3. Adapt a draft to a source you have inspected
These are original, unrendered planning examples. They assume a flat wall mirror with a plausible opening reflection, not a window, curved metal surface or rippling pool. Check your generator's actual model, input mode and duration before using either draft.
One small nod, with the mirror kept in view
Starting image: an adult faces a wall mirror. The camera is behind and slightly to one side, showing part of their head and shoulders directly and their face in the mirror. The mirror border is visible.
The person makes one small nod, then returns to the opening head position. The mirror shows the same person making that nod at the same time. Their clothing and hair remain consistent with the starting frame. The camera holds this view of the person and mirror.
Inspect the start of the nod, its lowest position and the return. The reflected face should not keep moving after the visible head pauses. Compare the action at the same playback time, rather than comparing two attractive frames chosen from different moments.
A slight head turn, with a defined ending
Starting image: the person and reflected face are already visible at an angle that leaves room for a small turn. The turn should not require revealing a face or mirror area absent from the source.
The person turns their head slightly to one side and pauses. The mirror shows that same head turn as it happens, with the reflection remaining part of the mirror view. The shot ends on the new paused pose. The camera keeps the opening composition.
Change “one side” to a direction you can describe unambiguously in your actual setup. Do not expect a screen-left movement in one view to prove the correct screen direction in another; viewpoint and mirror placement matter. Review the visible relationship instead of requiring a copied silhouette.
Both drafts deliberately leave out walking, hand gestures, dialogue and a moving camera. Those can be part of another brief, but each introduces additional movement to evaluate. If hands are essential, use the hand and grip checklist alongside the mirror review.
4. Compare the subject and reflection at matched times
Watch the full clip at its intended playback speed, then scrub the action and final pause. Note the time of a visible mismatch. Avoid guessing that a model internally “made a second character”; describe what the actual frames show.
| Check | Compare | Example observation |
|---|---|---|
| Appearance | Hair, clothing and other details visible in the two views | The reflected jacket changes color during the nod. |
| Placement | Reflection position relative to the person, mirror border and opening setup | The reflected head shifts across the mirror while the person stays in place. |
| Timing | Action onset, largest pose change and the final pause at matching times | The reflection nods after the person has already stopped. |
| Boundary | What belongs inside the mirror and what passes in front of its edge | A reflected shoulder appears outside the mirror frame. |
| Continuity | People, objects and the mirror surface across the entire take | An extra face appears in the mirror near the ending. |
Suggested acceptance checks, not measured model performance. On a narrow screen, scroll the table sideways; keyboard users can focus the region and use the arrow keys.
A frame with too little detail may not let you judge identity at all. Record that limitation instead of marking it correct. Likewise, a crop that hides the mirror cannot demonstrate a matching reflection. If the shot is meant to be intentionally uncanny, write that exception into the brief before assessing it.
5. Keep the next revision specific
Save the reference, exact prompt, model and mode, settings, output ID, mismatch time and the next change. If the starting arrangement is wrong, revise the source. If the mirror leaves the frame, revise the framing. If timing diverges during several simultaneous actions, try a smaller action that leaves both views readable, then inspect the new result without assuming the revision must work.
A successful take is not evidence of a universal mirror-lock phrase. If an accurate reflection is central to a demonstration or story, consider filming the shot or using a controlled compositing workflow. Removing the mirror or changing the edit can be a valid creative choice, but it changes the shot rather than repairing its reflection.
Draft the movement and ending in Shotmelo's image-to-video prompt planner, save the local draft, and export the text for your chosen service. The current preview does not render video or repair reflections. For identity changes beyond the mirror, consult the character consistency guide.
Common mirror-shot questions
Will “physically accurate reflection” keep it in sync?
It describes a goal, not a verified control. State the visible action and relationship, then inspect the result. Do not treat a good opening frame as proof that the reflection stays correct in motion.
Should the reflection copy every visible pixel?
No. The camera can see different sides of the person directly and through the mirror. Check the arrangement and movement expected in your setup, using features that are actually visible in each view.
Can I reuse these prompts for water or a shop window?
Adapt the shot and review rules first. Water can move and distort a reflection, while a window may show both reflections and the scene behind the glass. The stable flat-mirror assumptions in these examples do not describe every reflective surface.
Published November 8, 2026. The linked research abstract and Runway documentation were checked on that date. Drafts, observation card and checklist are original planning material, not rendered tests or claims of a successful reflection-generation method.