AI workflows
Turn a selected still into a Dreamina AI video shot
Begin a Dreamina AI shot with a still that clearly shows both partners and can support the intended action. Resolve image problems before adding motion, then ask for one small movement. Inspect the complete clip before expanding the scene so that a flawed starting image does not become an endless prompt-revision problem.
In this guide
- Select the stillBoth partners have clear roles and enough visible space for the intended movement.
- Resolve image issuesCheck identity, clothing, hands and the starting pose before animation.
- Request one actionDescribe a small movement and keep the camera instruction restrained.
- Inspect the resultWatch both people through the full clip and change one cause at a time.
Image preparation and video generation have different jobs
The still establishes the starting scene. Video generation introduces change over time. If Partner B already has an unclear face or an arm that blends into the background, asking for a gentle embrace does not remove the uncertainty in that starting material. Inspect the image before deciding whether the motion prompt needs more detail.
Dreamina's basic Help covers creation and review, while its image-to-video resources describe image-prompt workflows. Those documents support planning a shot from an image; their existence does not verify that every two-person interaction will preserve both identities. Treat product examples as demonstrations of a workflow and your own result as the thing that must pass review.
The Zombie Couple site currently offers a local two-photo casting preview and plays its reference films. Generation and checkout are disabled. Dreamina AI is an independent tool discussed here for planning an external experiment; the steps below are not controls available inside this site's preview.
Choose a still that can support the proposed action
Use a quiet memory shot as the first still-led exercise: two partners seated on a bench, with a light breeze moving a scarf and loose hair. Before requesting animation, inspect whether the still actually shows a readable bench, separate bodies and the scarf edge. A crop that hides the second person’s shoulder or already fuses the scarf with a hand gives the motion an uncertain starting shape.
Record a simple approval condition for the starting still: both faces recognizable, clothing separated from the bench, no unexplained extra hand and enough room around the fabric that should move. These are your review conditions, not Dreamina command syntax. The photo guide covers selection of original portraits; this pass evaluates the composed image that will begin the shot.
For portrait selection and local role assignment, use the existing photo guide. Here the question is narrower: can this composed still begin this shot? Keep an unchanged copy of the selected image so later prompt comparisons have the same starting point.
Make an image approval pass before adding motion
Review each face separately, then inspect the relationship between the bodies. Check whether an arm belongs to the correct person and whether a shoulder disappears into the other partner. Look for clothing details that will matter in the next shot. A still that looks acceptable at thumbnail size may reveal a problem when you inspect the hands or the farther face.
If the project requires a zombie state and a healthy memory state, label them as separate visual states. Decide which state this shot begins in. Asking for identity, wardrobe, health state and camera direction to change at once creates several moving targets. A simpler first experiment holds the starting appearance steady and changes only the action.
- Compare both faces with the intended people at a useful viewing size.
- Trace visible arms from shoulder to hand, especially where the partners overlap.
- Record the starting story state and clothing that should remain stable.
- Keep the source still unchanged for the first motion comparison.
Create and download the first video attempt
Dreamina’s current Help describes this route: sign in, choose video generation, write the requested movement or add a reference image where supported, choose the available settings, then generate. Review the result before downloading it. Record the selected model and settings shown in your account; the help does not promise identical controls or allowances for every account.
For the bench exercise, use the approved composed still as the image input when the selected operation supports that role. Check whether the upload is a starting image or a general reference. If the interface offers start and end frames, those slots describe temporal endpoints; they are not automatically separate identity slots for the two people. Do not attach unrelated headshots as if they were a complete beginning and ending scene.
Keep the initial image and downloaded attempt under distinct names so you can compare them later. A local review copy is useful even when the result fails, because you can point to the moment the scarf changes shape instead of relying on a thumbnail. Confirm any export or account restriction shown before committing further attempts; no fixed credit price is assumed here.
Write the action in a short sequence
Use the written request to add motion that the approved still can support. In this exercise, the people remain seated while one visible environmental element moves. This gives you a specific comparison between the static starting material and its animated interpretation, without combining an approach, a hug and a camera move in one attempt.
The example below is an original proposed test, not a Dreamina output. If the scarf is already tangled with an arm in the starting still, repair or replace that still before testing fabric movement. Extra prompt adjectives cannot supply a clear original edge. The separate couple-prompts guide covers general wording; here the useful decision is whether the image is ready to animate.
Proposed bench-memory motion
Both partners remain seated side by side in the position shown. A light breeze gently lifts the loose end of the scarf, then lets it settle. Keep the bench and camera view steady, with both faces visible. End after the fabric settles. This describes one motion to inspect against the approved still.
Review identity through the movement
Watch the clip once at normal speed and decide whether the bench scene reads as the same quiet moment. Then compare each face at the beginning, during the scarf movement and at the end. Check the scarf edge against the original image: does it lift from the correct shoulder and settle without becoming part of the other person’s clothing?
For body contact, pause around the first overlap. Track the arms from the starting pose into contact and through the final hold. A hand can be hidden naturally, but it should not reappear attached to the other person or change the body arrangement without a visible cause. Note the first frame where your confidence in the movement breaks.
Also check the background and camera. If the request used a fixed view but the frame moves enough to hide a partner, the result misses the review condition even if the surviving face is convincing. Evaluate the shot against the written brief rather than accepting an unrelated attractive moment.
Revise one detail, then decide whether to keep the shot
Dreamina's Help recommends refining details rather than treating the first result as final. Use that principle to isolate a cause. If the still is unclear, return to image preparation. If the movement is too ambitious, simplify the action. If a camera move obscures the second partner, hold the view steady. Keep a short record of the change and the defect it is intended to address.
If the scarf starts as a clean shape but splits into extra fabric during motion, retain the same still and request a smaller movement in the next attempt. If it was already merged with a hand before animation, correct the still instead. These are separate diagnoses. Keep the unmodified first result so the later comparison shows which change addressed the visible defect.
Stop when the result fails the same essential condition and you no longer have a specific revision to test. Save the failed attempt's notes instead of adding unrelated adjectives. When a shot passes, retain its still, prompt and role record before moving to the memory scene. The next shot inherits a continuity problem of its own, even when the first one works.