Temporal consistency asks whether a shirt stays the same colour from frame to frame. Physics consistency asks whether the shirt hangs like cloth when the wearer turns. They fail independently. A clip can be flicker-free and still make a bottle slide through a table, a car take a corner with no lean, a person plant a foot that does not take weight. Kling 3.0's February 2026 pitch leaned on this axis — native 4K plus better physics and camera motion — because the previous generation's tell was not resolution, it was mass.
The usual breaks are contact, conservation, and secondary motion. Contact: hands that hover a millimetre off the product, feet that skate. Conservation: objects that change size as they move, liquid that does not pour. Secondary motion: hair, cloth, and camera shake that do not follow the primary action. Prompting 'realistic physics' does almost nothing. Showing the model a real motion (a driving clip, a first-last pair that is actually connectable, a reference of the object in the hand) does more.
Native 4K will not fix a physics miss. It will make the miss sharper. Camera-control sliders that request a move the mass cannot survive will invent a warp. When a shot has to sell a product in a hand, film the hand or generate the orbit around a still you own rather than asking the model to invent a grip.
In practice
- Review at full speed, then at the contact frames — the handshake, the pour, the footfall.
- If the object must be held, use a real hand or a still you own. Invented grips are where physics dies first.
- Do not 'fix' a floaty object by upscaling. More pixels will not add mass.
The mistake to avoid
Reading a physics miss as a resolution problem. 4K on a bottle that phases through the table is a clearer ghost.
Related terms
Temporal consistency
Temporal consistency is how well a generated clip keeps things the same from one frame to the next, with no flicker, shimmer or drifting detail.
Motion level
Motion level is a coarse dial for how much movement a generated clip contains, usually offered as low, medium or high rather than a precise number.
Camera control
Camera control is the set of inputs that decide how the virtual camera behaves during a generated clip — whether it pushes in, orbits, pans, or stays locked off.
Prompt adherence
Prompt adherence meaning: how faithfully a model does what the prompt said, down to the count, position and colour, not just something attractive nearby.
Character consistency
Character consistency meaning: whether the same person, mascot, or product still looks like itself across separate generations.
The all-in-one AI studio for creators. 60+ models for video, image, voice, music and lipsync, in your browser or on your phone.
Free on the web and on iPhone, one account.