AI video prompt generator. Ten decisions, one clean shot.

A weak video prompt fails in a way an image prompt doesn’t: the pack melts halfway, a sixth finger appears, the shot cuts to a scene nobody asked for. Pocket Studio · Video walks you through the ten decisions a director makes before calling “action”, then writes the prompt for Kling, Veo (in Gemini or Flow) or Sora, with the must-never-appear list built in.

It’s free, there’s no sign-up, and what you type stays in your browser.

How it works

1. Decide. Ten short screens, one decision each. The counter at the top tells you how many you’ve left to the model.
2. Paste. Pick Kling, Veo or Sora. The prompt changes shape for each one, and a small box tells you which settings belong in the tool, not in the prompt: ratio, length, sound.
3. Fix one thing. If the clip is almost right, the “one fix” tab gives you a single line for the single problem: morphing hands, a warped logo, flicker, a surprise cut.

The prompt you copy is always in English. Kling, Veo and Sora follow English best, and English is the prompt language Google documents for Veo.

The ten decisions behind a professional video prompt

I teach this order in my workshop. People arrive with one line, “a nice video of my product”, and leave knowing why the model gave them somebody else’s product. Skip a decision and the model takes it for you, quietly:

1. Format and length. Reels and Stories 9:16, YouTube 16:9, square, or an Instagram post. None of the three models outputs 4:5, so for a post the tool frames a vertical shot that survives the crop. About five seconds or about ten.
2. Subject and one action. What is in the frame, and the one thing that moves. Two actions in five seconds is two shots.
3. Shot size. Close-up, medium, wide or a single detail.
4. Camera movement. Locked-off, slow push-in, pull-out, pan, orbit, tracking, crane up or handheld. One move per shot; a push-in and an orbit in five seconds is how a pack starts to bend.
5. Speed. Real time, slow motion or timelapse.
6. Light. Window light, soft studio, golden hour, night with practical lamps, or backlight.
7. Setting. A plain backdrop, a tabletop, a real interior, outdoors or a city street.
8. Style. Cinematic live action, soft 3D, 2D animation, stop-motion clay or documentary realism.
9. Mood, colour and sound. What the viewer feels, up to three colours, and whether you want ambient sound, one short line of dialogue, or silence so you can add sound in the edit.
10. Start frame. From text, or from an image you already have. With an image, the prompt describes only motion and camera. The picture already shows the rest.

Six things are always on the never list: morphing or melting hands and faces, extra fingers, warped or changing logos and pack text, flicker, sudden cuts, and text appearing on screen. Kling gets them as a separate negative prompt. Veo and Sora get them as plain “avoid” sentences, because neither the Gemini app nor Sora gives you a negative field.

Kling, Veo or Sora: what changes in the prompt?

Those are the facts I could check in each model’s own documentation in October 2026. Where a fact couldn’t be checked, the tool leaves it out instead of guessing.

Before and after: same idea, two prompts

What most people type:

a nice video of a coffee cup for instagram reels

What Pocket Studio · Video writes for Kling after ten choices:

Subject: A ceramic cup of coffee with latte art on a wooden table.
Action: Steam rises slowly from the cup while a spoon is set down beside it. Everything moves at real-time speed.
Scene: The setting is a simple tabletop; the background is softly out of focus.
Camera: Close-up: the subject fills most of the frame. The camera slowly pushes in toward the subject (dolly in). One continuous shot, no cuts.
Lighting: Soft natural window light from one side, with gentle shadows.
Style: Cinematic live-action footage, as if shot on a real film camera.
Mood: Warm and cosy. Colour palette: only cream and brown.
Sound: quiet café murmur and the soft clink of a spoon on porcelain. No music, no dialogue.
Timing: The shot lasts about 5 seconds, and the action fills the whole shot.

And the negative prompt that goes in Kling’s second box:

morphing, melting hands, distorted face, extra fingers, deformed hands, warped logo, changing packaging text, distorted label, flicker, flickering light, jump cut, scene change, transition, on-screen text, subtitles, captions, watermark

The first one leaves every decision to the model, including the ones that ruin a product shot. The second leaves none.

Three shots that cut together

One good clip can be luck. Three that play as one film is direction. In “A set of 3 consistent shots” you lock the light, colours, setting and style once, choose how the set moves (one scene in three shot sizes; one subject, three camera moves; before, during, after) and add one small detail that repeats. The look lines are word-for-word identical in all three prompts. That, plus a still from shot one where your tool takes a reference image, is what holds the set together. The models don’t remember shot one on their own.

When a prompt isn’t enough

If the pack on screen has to be your pack, start from a real photo with “I have an image”. The prompt then adds only motion and camera, and tells the model to keep the shape, colours, logo and pack text. Even then I throw away 60 to 70 percent of takes. The method doesn’t make the model perfect; it makes each rejected take cheaper and each retry deliberate.

Need still images first? Use the sibling tool, the AI image prompt generator. For a campaign or a launch film, that’s my day job: AI commercials for brands, and how I work. The method comes from my workshop, which runs in Persian: the workshop page.

FAQ

Built by Mohammad Nasravi, AI creative director. Model facts checked in each model’s own documentation, October 2026.