AI VIDEO
PROMPT BUILDER.
Build a structured shot description for Google Flow, Veo, Runway, Sora, Kling or Luma — subject, action, camera, light, framing. Runs in your browser, no signup.
Fill in the fields and click “Generate”
If you are animating from a still, the seed decides more than the prompt does. We ran a controlled A/B — same model, same wording, only the seed changed — and the output flipped from a re-composed frame to held framing and correct physics.
- 1. Seed the state you want to animate from, not the finished product. A shot that already shows the end has no motion left to perform.
- 2. One subject, on a real background. Composites on transparency invite the model to re-compose the frame.
Subject and action carry the meaning. Camera, light and framing qualify it. Leave a field on none rather than filling it for the sake of it — contradictory directions are the most common cause of a wasted generation.
The prompt body is the same everywhere, because these are cinematography terms, not product features. The target only changes the technical note about duration and where to set aspect ratio.
If you are animating a still, iterate the seed image before you iterate video. Image generation is several times cheaper than video, and a failed generation bills the same as a successful one.
On slash shorthand: the slash tokens are community convention, not official commands in any of these products. The model reads the word, not the symbol — so they are a useful way to keep a batch of prompts consistent, and no substitute for describing the shot. We keep them optional for that reason.
FAQ.
What makes a good AI video prompt?
Lead with subject and action — what is in frame and what it does — then qualify with camera, lighting and framing. Describe the shot in sentences rather than listing keywords, because these models read language, not tags. One clear action beats three competing ones.
Do slash commands like /cinematic actually work?
They are community shorthand, not official commands in Google Flow, Veo, Runway, Sora or anywhere else. The model reads the word, not the slash. That does not make them useless — they are an efficient way to keep a batch of prompts consistent — but no token switches on a hidden feature, and a list of a hundred of them is a glossary, not a control panel.
Will the same prompt work on Sora, Veo, Runway and Kling?
The body transfers, because subject, camera, lighting and framing are cinematography vocabulary rather than product features. What differs is tolerance and length: Sora handles longer narrative descriptions, Runway rewards concision, and each has its own clip duration and aspect ratio settings. That is why this tool keeps one prompt body and changes only the technical note.
Why does my AI video flicker, or the subject keep changing?
Usually too much simultaneous movement. Reduce it to one or two moving elements, and if the background should hold still, say so explicitly — an unstated background is one the model feels free to animate. Consistency problems are more often an over-specified prompt than a weak model.
Why does the model ignore my framing and re-compose the shot?
If you are animating from a still image, the seed decides more than the prompt does. We ran a controlled A/B — same model, same wording, only the seed changed — and the output went from a re-composed frame with an invented background to held framing with correct physics. Use a single subject on a real background, and seed the state you want to animate from rather than the finished result.
How long should an AI video prompt be?
Precise beats long. A specific thirty-word prompt usually outperforms a vague hundred-word one, because every extra clause is another instruction the model has to reconcile. When a result is close but wrong, change one element and regenerate rather than rewriting the whole thing — otherwise you cannot tell which change did the work.
Is this tool free, and is anything stored?
Free, no signup, and nothing is stored or sent anywhere. The prompt is assembled in your browser from the fields you fill in.