Video generation is usually approached backwards, the way a traveller might book the return flight before choosing a destination. A scene is briefed in text, the expensive dice are rolled, and whatever woman the render invents is accepted as the star. The stills of the series must then compete with a stranger’s clip. Keyframe-first reverses the order of travel. The still is made first — on character, already judged, already logged — and video becomes a motion pass over an arrival that has already been approved.
The four legs
1. Make the still as a first frame. Compose with movement in mind: the traveller placed with room around her, a direction left open to turn or walk toward, a background that rewards a few seconds of looking. A still composed as a poster makes a cramped keyframe. Leave the frame somewhere to go.
2. Brief one movement. The motion line is a stage direction, not a plot: “turns slowly toward the river light and lowers her eyes.” One movement for the traveller. None for the camera. Every added moving element multiplies the ways a render can smear, and a short clip is read in a glance anyway.
3. Render at the destination’s measure. Video pricing follows duration and resolution, so decide where the clip will live before it is rendered. A feed teaser and a store piece are different journeys with different fares. Rendering the maximum of everything, just in case, is how a travel budget quietly doubles.
4. Judge it as a clip. Watch at full speed, twice. Once for the traveller: does she stay herself through the movement? Once for the joins: hands, hair, the edges of the background at the turn. A clip that passes both viewings joins the set. One that fails receives a new motion line — not a second roll of the same one.
Why the planned route costs less
The expensive mistakes in video are character mistakes, and keyframe-first moves them back to the 6-credit still, where they cost a fraction to correct. By the time a render is priced in video credits, the only open question is motion — which is the question video models answer best.