lolq.ai

← All lessons · 6 min read Читать по-русски

Motion: what the engine will not ruin

Two complaints about video sound like opposites and have the same cause.

"Nothing moves." The motion prompt describes a mood, not an action. atmospheric, cinematic, dynamic is zero instructions. The engine has nothing to do and honestly does almost nothing.

"Everything smeared." Three actions were put in one scene: the character turns, the camera travels, a car passes behind them. The engine tries all three and delivers none.

One rule covers both: one movement per scene.

Who is moving

Before writing the motion prompt, answer one question — what exactly travels:

The third option is the most underrated. A scene where neither the camera nor the person moves is still alive — and it never falls apart, because the engine never has to preserve anatomy through movement.

Five scenes that do not fall apart

DeviceWhat movesWhy it comes out clean
The turnsubjectone body part rotates
Running at the camerasubjectmovement on one axis, no rotation
Cloth in the windenvironmentfabric is easy, the face is untouched
The pass-byenvironmentone large object crosses, everything else is static
Hands, no facesubjectthere is no face in frame to get wrong

All five are open on the free plan. Four of the five are deliberately built so the engine gets exactly one job.

A note on "Hands, no face"

This is the most underrated card in the catalogue. Every clip has scenes where the face drifts, and they are usually fixed by regenerating five times. Hands cannot drift: there is no face in the frame to ruin. The emotion survives anyway — two hands doing one small thing read more precisely than a bad face.

A note on "The pass-by"

A car crosses the frame and briefly hides everything behind it. In the last frame there is a person standing where nobody was in the first. It is a free wipe: a change of location inside one scene, with no cut and no editing software.

Steps

  1. Walk the storyboard and find the scenes whose frame animation prompt field

contains adjectives. Those are your future dead frames.

  1. Decide for each one what moves: camera, subject or environment.
  2. Apply the pack to five scenes and fill the slots: character, location,

outfit. The outfit matters — it is what carries recognition between scenes.

  1. Rewrite the motion on the remaining scenes on the same principle: one verb.
  2. Generate frames, get the pictures right, and only then the video.

What to look at

Check the video of the first scene before animating the other twenty-nine. Regenerating video is the only action in the studio that costs credits every single time. The order "one scene → look → the rest" saves more than any saving on engines.

Watch the edges of the frame. If a hand has drifted or the face has become a different person, the problem is not the animation — it is that too much is moving in that scene.

The usual mistakes

Adjectives instead of verbs. dynamic, energetic, cinematic motion carries no information. The camera moves forward on one axis, the subject stays still is an instruction.

Two verbs in one scene. "Turns and walks" will be read as neither. Split it into two scenes — they will sit better against the music anyway.

A push-in and a turn at once. The classic. The camera travels forward while the character rotates, and both smear. Keep one.

Regenerating video instead of regenerating frames. If the problem is in the picture, video will not fix it but will charge for it. Redrawing frames is free; regenerating video is not.

Twelve-second scenes. The longer the scene, the more the engine has to invent. A long scene needs mid-frames, otherwise it picks the shortest, dullest path between the start and the end.

The artefact

Pack "Motion that survives a cheap engine" (motion_basics) — five scenes built so the engine gets one job: the turn, running at the camera, cloth in the wind, the pass-by, hands only. Slots: character, location, outfit. Free plan.


Next: Editing: half the clip lives at the cut.