Text-to-Video Prompts That Work
A video prompt is a shot description, not a story. Describe one shot well and the model will render it; describe a scene and it will guess.
18+ only · fictional characters only · no public gallery
The short version
Three decisions that matter more than anything else.
Name the subject
Who or what is on screen, in a few concrete words: age range, clothing, material, colour. Vague subjects produce generic frames.
Describe one action
One verb that can happen in a few seconds: turns, walks toward the camera, steam rises. Several actions in one clip compete and blur.
Set camera and light
Shot size, camera movement and lighting: close-up, slow push-in, soft window light. This is what makes it look filmed rather than drawn.
The four parts of a good prompt
Subject, action, camera, light. Almost every prompt that works contains those four, and almost every prompt that fails is missing one. A useful order is the order a director would give it: what we see, what it does, how we look at it, how it is lit.
Example: a woman in a red raincoat on a wet city street at night, she turns toward the camera and smiles, medium close-up, slow push-in, neon reflections, shallow depth of field. Every phrase there changes the picture; nothing is decoration.
What to leave out
Leave out the plot. A clip lasts a few seconds, so anything that needs a before and an after belongs in a second clip. Leave out abstract adjectives such as beautiful, epic or amazing: they do not describe anything the model can render.
Leave out negative instructions where you can. Writing no blur puts blur into the description. Describe the sharp thing you want instead.
Example prompts by shot type
Portrait: close-up of an elderly man with a grey beard, he blinks and looks out of the window, soft morning light, static camera. Product: a glass perfume bottle on black stone, slow orbit around it, single hard light from the left. Landscape: fog over a pine forest at sunrise, slow aerial drift forward, warm low sun.
Each of these is one subject, one action and one camera move. Run them in a text to video generator and change one part at a time, so you learn which word moved the result.
When a picture beats a prompt
Text is the right tool when you know the idea but not the exact look. When the look is already fixed — a specific face, a product photo, a character you designed — a prompt will keep drifting away from it. Start from the picture instead and use an image to video generator, which keeps the frame and only adds motion.
For recurring characters across many clips, consistency matters more than any single prompt. A character video generator built around one reference face will hold it better than re-describing her in every prompt.
Frequently asked questions
How long should a text-to-video prompt be?
One or two sentences, around 20–40 words. Longer prompts tend to contain competing instructions.
Should I describe several actions?
No. One action per clip. Split a sequence into several clips and join them afterwards.
Why does the result ignore part of my prompt?
Usually the prompt asks for too much. Remove everything except subject, action, camera and light, then add back one detail at a time.
Do negative prompts help?
Rarely. Describe what you want rather than what you do not want.
Can I get the same character in every clip from text alone?
Not reliably. Start from a reference image when the character has to stay the same.
Try it on your own image
New accounts get starter credits. No card needed to run a first generation.
Try it yourself