A strong nonhuman transformation and faithful camera motion are separate requirements. Luma's Ray3 Modify guide explains that high Reimagine settings can loosen shape and background constraints and reinterpret camera movement. If the character becomes convincing while the shot unexpectedly pushes in, do not call the whole result accurate because the creature design worked.
Define the camera evidence in the original
Watch the source without thinking about the new character. Identify stable background features, the subject's screen size and any movement that genuinely belongs to the camera. A performer leaning toward a fixed lens is not the same as a camera dolly. Record what can be established from the image, and leave uncertain motion described cautiously.
Choose landmarks that remain visible long enough for comparison: the edge of a doorway, a wall seam or a fixed tabletop corner. A moving sleeve or another walking person is a poor camera reference. Avoid using only the transformed subject's silhouette, because changing that silhouette is precisely the requested task.
Specify two independent passes
The character pass can require an approved creature shape, readable action and consistent identity. The camera pass can require the original framing progression without a new push, pan or background shift. Write what degree of variation is acceptable before looking at a favored output.
If the shot permits some reframing, state that explicitly. Otherwise the reviewer may excuse an unwanted zoom simply because it helps hide difficult anatomy. Aesthetic preference is a legitimate separate decision, but it should not silently rewrite an original promise to preserve the camera.
A hypothetical strength comparison
Use the same authorized source and transformation goal for a small supported comparison. Label settings by the actual values or names exposed in the current mode rather than inventing a universal numeric scale. Keep the attempt allowance explicit; this guide has not executed or measured these trials.
| Candidate | Character observation | Camera observation | Decision |
|---|---|---|---|
| More adherent | Does the intended creature form emerge? | Do source landmarks stay consistent? | Insufficient change or useful fidelity? |
| More reimagined | Does anatomy remain readable? | Is there an unrequested push or shift? | Accept both criteria or reject? |
Match source and result times before comparing landmark positions. If output duration changes, document that first. A two-frame timing error during genuine camera movement can produce a misleading displacement. Record random-variation limitations instead of declaring one setting generally best from one attractive result.
Choose a viable shot, not the largest setting
If no tested candidate satisfies both requirements, revise the production plan. A closer match between performer proportions and the intended creature may reduce the transformation burden. A simpler camera setup or a deliberate new shot can also be more honest than claiming a changed perspective is preserved motion.
Inspect the complete movement and any cut into the next scene, then retain the accepted candidate and its settings with the camera evidence. Luma's strength behavior is specific to its documented workflow; Imagild does not inherit that control or guarantee the same result. For an Imagild brief, record both the desired transformation and protected camera relationship. Completion requires approving both, or explicitly accepting a revised shot concept.