Text-to-video prompts for Sora, Runway, Pika and Vidu

The default structure for a single generated shot

[SHOT SIZE] of [SUBJECT] [DOING WHAT], [CAMERA MOVEMENT], shot on [LENS/FORMAT], [LIGHTING], [TIME OF DAY], [MOOD], [COLOR PALETTE], [SECONDS]-second continuous take, no cuts, no text, no people looking at camera.

In practice: Wide aerial shot of a container ship cutting through calm water, slow forward drone push, shot on a 35mm anamorphic lens, low golden hour side light, dawn, quiet and vast, muted blue and amber palette, 5-second continuous take, no cuts.

Converting narration into generatable clips

Here are lines from my video script: [PASTE LINES]

For each line, write a text-to-video prompt using this structure: shot size, subject, subject motion, camera movement, lens, lighting, mood, palette, duration, negative constraints.

Rules: no dialogue, no on-screen text, no recognisable public figures, no logos or brands, no complex hand movements, no more than one subject per shot. If a line cannot be shown without those, say so and suggest a visual metaphor instead.

In practice: The 'cannot be shown' flag saves the credits you would burn on impossible shots.

Multiple clips that belong to the same video

I need [N] clips for one sequence. Define a fixed style block — lens, film stock or grade, lighting logic, palette, and grain — that will be appended verbatim to every prompt.

Then write [N] prompts that vary only subject, motion, and shot size, each ending with that identical style block.

Subjects: [LIST]

In practice: Style drift between shots is the fastest way to make generated b-roll look generated.

Animating a still you already like

Animate this still image. Motion: [SPECIFIC MOTION — e.g. slow parallax push in, dust drifting through the light shaft, water rippling in the lower third]. Keep everything else static. No camera roll, no zoom out, no new objects entering frame, no morphing of the subject. [SECONDS]-second loop.

In practice: Naming what must stay still matters more than naming what should move.

Fixing warped, flickering, or morphing output

This prompt produced [DESCRIBE THE FAILURE — warped hands, flickering background, subject morphing mid-shot, camera drifting]. Prompt used: [PASTE PROMPT].

Rewrite it to avoid that failure mode. Reduce the number of simultaneous motions, simplify the subject, shorten the duration if needed, and add explicit negative constraints. Explain which change targets which failure.

In practice: Most failures are too many simultaneous motions in one shot. Split it into two clips.