Compare a footage transformation in Luma Dream Machine

Choose a performance worth preserving

Start by watching the unmodified clip. For this example, one partner waits beside a station window while the other enters carrying a small bag and stops nearby. The test should change the arriving character's fictional appearance while retaining enough of the entrance and stopping action to preserve the scene's meaning.

Use footage you are entitled to modify and people whose likenesses you have permission to use. The reference films on Zombie Couple can help you study a story beat; their availability is not permission to recast and republish them. A self-recorded practice clip makes the performance and permissions easier to document.

Write the preservation brief before opening the modification controls. In this example, preserve the entrance direction, the pause beside the window and the waiting partner's role. Appearance is the intended change. If a result makes the bag disappear but keeps the face, you need to decide whether that breaks the action, rather than judging the face alone.

Match the procedure to Ray3 Modify

The official Ray3 Modify guide, published December 12, 2025, separates source video, character reference, keyframes and strength. Its basic route uploads a source clip and character image in Modify mode, then adjusts strength before submission. This article applies to that documented workflow, not every Luma model.

For its keyframe route, the guide opens a frame card in the Modify frame editor, enables the character option and submits a frame edit. Repeat for another frame if needed, then set strength for the video. It warns that a clip longer than five seconds samples the five-second position for the displayed end frame, while transformation still applies throughout.

Treat those controls as separate decisions in your test sheet. Loading a source identifies the performance. Accepting a reference identifies the proposed appearance. Approving a keyframe identifies a desired frame state. Selecting strength identifies how broadly you want to test the transformation. None is a substitute for watching the completed moving clip.

Mark matching moments in the original

Make a short register before generating anything. Choose moments that explain the action: the arrival begins, the bag crosses the body, the character stops and the waiting partner reacts. Describe what is visible at each point. This gives the modified result a comparison structure even if its timing shifts slightly.

Record the original camera behavior. If the source is fixed, note which edges remain stationary. If it moves, note how the window and doorway move through the frame. These are visible landmarks for checking whether a later transformation changes framing enough to alter the performance.

Include a moment where the arriving character's face is partly turned. A reference might look convincing in a front view while becoming unfamiliar during the turn. Also include the waiting partner even when they are not the intended target; unintended changes to that person can break the scene.

Inspect keyframe edits before they become motion targets

If you choose the keyframe route, review each edited frame as an image. Does the intended character occupy the same role? Are the body pose and bag placement usable? A strong facial change alongside a broken shoulder is not an approved frame for this experiment. Save the comparison notes before moving on.

Use the documented sampled-frame caveat to decide what you can infer. A frame labeled as an ending state may represent an earlier moment in a longer source. Record the actual source position rather than assuming it proves the last moment of the clip. The final portion still needs a separate moving-result check.

Avoid combining unrelated changes while approving the frames. If the purpose is a character appearance test, keep the waiting room and bag requirement stable. A new room, missing prop and changed pose introduce other explanations for the final result. You can plan a larger fictional transformation after the narrower comparison teaches you what changed.

Treat strength as a tradeoff to inspect

The documented slider runs from Adhere toward Reimagine: closer adherence favors source geometry, while stronger reimagining permits broader changes. The guide also warns that stronger settings can reinterpret camera movement. These are reasons to compare, not promises that a particular position will preserve your scene.

Choose an initial position and write it down. After the first result, decide whether the appearance changed enough and whether the movement remained suitable. If you test another strength, keep the same source, reference and accepted frame edits. Otherwise you cannot tell whether the difference came from strength or a changed input.

Do not call the strongest transformation the best merely because it looks different. In the station example, a broader redesign could erase the recognizable arrival or redirect attention away from the waiting partner. The preferred result is the one that meets your brief, even if its visual change is less dramatic.

Compare performance, appearance and framing separately

First watch the original and modified clips in sequence at normal speed. Does the same action read clearly? Then compare the registered moments. A synchronized side-by-side review can help, but only if you account for any timing shift; matching timestamps may no longer represent the same gesture.

Inspect the intended character's face and body, the bag and the waiting partner separately. Record each result as acceptable, uncertain or failed. An uncertain concealed hand is different from a clearly misplaced hand. Keep uncertainty visible rather than writing a blanket “everything preserved” judgment.

Check the camera last against your recorded landmarks. A new zoom can conceal an altered prop or remove the second person from view. If the framing changes, decide whether the film can use that composition. You may accept it as a new shot, but it no longer demonstrates preservation of the original framing.

• Compare the arrival and stopping gesture at ordinary playback speed.
• Inspect the target identity during its turn, not only at the start.
• Check the bag, arms and waiting partner for unintended changes.
• Inspect the final portion beyond any sampled keyframe.
• Record camera changes separately from character changes.

Keep the modification only when the comparison supports it

Choose among a clear keep, one specific revision or a stop. If the face works but the bag no longer explains the action, the shot may need a narrower edit or a different source. If the camera has changed yet the scene still works, record that as an accepted creative difference rather than a preserved property.

Retain the original, selected reference, frame-edit decisions and strength setting with the chosen result. Those records support a later continuity check and prevent an attractive variant from becoming the new baseline without explanation.

Zombie Couple currently offers local casting preview and reference-film playback; generation and checkout are disabled. This Luma Dream Machine guide describes an independent documented operation and an original comparison method. No paid render or output-preservation test was performed.