Camera moves in metres and seconds, not adjectives

You type ‘slow push-in’ and the generated clip barely moves at all. The model hasn't failed: you got exactly what you asked for. To a video model, slow means little movement, and in a four-second clip little movement easily rounds down to none.

The fix is to give size and time instead of adjectives: where the camera starts, where it ends up, and how long it takes. This article gives you the formula, with prompt fragments you can paste today.

Why ‘slow’ freezes the shot

Our own planner keeps these words on a blocklist for video prompts: slow, gentle, subtle, slight, barely, calm, soft, minimal. Each of them tells the model the same thing: better not to move. In practice, that is what freezes the picture.

The other problem is that slow is relative. Slow compared with what? The model has no yardstick, so it falls back on the safe, static average. A number leaves no room for interpretation: one metre is one metre, five seconds is five seconds.

The formula: from where, to where, over how long

A well-written camera move has four parts. With all four in place, the model has nothing left to guess.

1. One moveName a single move: push-in, pull-back, quarter arc, tilt up. If you need two, that is two shots.
2. First frameSay what we see as the move begins: the framing, and where the subject sits in the frame.
3. Last frameSay where the camera ends up. This is exactly what ‘slow push-in’ never tells the model.
4. Distance and timeGive the distance (as framing or in metres) and the time in seconds, and state that the move is clearly visible and continuous.

The camera moves in from a medium shot to a close-up of her face over the full four seconds, clearly visible and continuous. The shot opens with her head and shoulders in the lower half of the frame and ends with her eyes on the upper third line.

Distance two ways: framing or metres

When moving towards a face, framing is the safest unit: ‘from a medium shot to a close-up’. The model reads it as a change in the picture, which makes the result easy to check.

For moves that travel through space (along a counter, down a street, round an object), metres and centimetres work better. In a short clip, a few tens of centimetres read clearly next to a close object, while a landscape needs several metres for the same effect.

The camera travels about one metre forward along the bakery counter over five seconds, at a steady, continuous pace, starting at the edge of the bread basket and ending with the baker's flour-dusted hands filling the frame.

Camera and character in separate sentences

Squeeze camera and character into one sentence and the model may mix up what is moving. So the camera move gets the first sentence, and the character's behaviour gets the next ones.

Before: ‘slow emotional push-in on a sad man by his bike’. That has a banned word, an abstract emotion, and no distance or time. After, it reads like this:

The camera pulls back about two metres over five seconds, from a close-up of his hands on the bicycle handlebar to a wide shot of the empty street. He exhales and his shoulders drop. Wind moves the loose hem of his jacket, and a passer-by crosses behind him on the left.

The sadness has become a visible sign: he breathes out, his shoulders drop. The wind and the passer-by keep the shot alive even when the camera pauses for a beat.

Each model wants a different frame

Kling likes concise, natural full sentences (‘The camera moves…’) with one move. Negatives go in their own field, never in the prompt. Seedance expects 60 to 100 words (200 is the hard ceiling) and takes no negative list. Veo follows the same principles, kept concise.

For talking shots, keep the camera moderate at most: a push-in or a quarter arc. Add any handheld feel later, in the edit.

Before you paste: a quick check

• Search for slow, gentle, subtle, soft, calm and slight, and swap each one for a number.
• Is there a first and a last frame? If only the direction is there, add where the camera ends up.
• Does the time match the shot length? A five-second shot gets ‘over five seconds’.
• Is there only one move? A second move means a new shot.
• Are camera and character in separate sentences?

Take the last prompt that gave you a frozen clip. Strike out the adjective, write in a distance and a time, and generate again from the same start frame. You will see the difference on the first try.