Text to video prompts that work: 12 examples you can copy
A text to video prompt works when it names what moves, not just what is in the frame. Here is the subject-camera-motion-light structure, the dials that fix a prompt without rewriting it, and 12 copy-ready prompts for Indian businesses.
· 6 min read · 536 views · by the Cortex Digital Hub team
Paste one of the 12 prompts below, pick the model and length suggested, and see the credits before you press Generate. Draft on a cheap model, then spend on the final. From ₹24 a clip · no watermark · runs in your browser
Make a video → → Opens AI Video Generator · runs in your browser
A text to video prompt is not a caption. A caption says what is in the frame; a prompt has to say what happens in it. Most disappointing clips come from prompts that would make fine photo descriptions, and the fix is the same every time: name the movement. Below is the structure that works on the AI Video Generator, the dials you can turn when a prompt is close but not right, and 12 prompts for Indian use cases that you can paste as they are.
Say what moves, not just what is there
This is the site's own rule, and it covers most of what matters. "A bowl of ramen on a table, soft light" is a photo. "Slow dolly in on a bowl of ramen, steam rising, soft window light, shallow depth of field" is a shot. The second one tells the model three things the first does not: the camera is moving (dolly in), something in the scene is moving (steam), and how fast (slow).
A video model is predicting motion. If you do not give it any, it will invent some, and what it invents is rarely what you would have chosen. Flags flutter when you wanted them still, a hand appears from nowhere, the camera wanders. Writing the motion yourself closes off those guesses.
The four parts: subject, camera, motion, light
A prompt that behaves usually has four parts, in roughly this order.
Subject is the one thing the clip is about, described in a few concrete words: "a glass jar of mango pickle with a white label", not "a product". Camera is what the lens does: static, push in, pull back, slide left, tilt up, orbit. Motion is what happens inside the frame: steam rises, fabric ripples, a hand lifts a spoon, petals drift. Light is the mood: warm window light, overcast soft light, neon at night, golden hour.
You do not need all four every time. A static camera is a valid choice and often the best one for a product. But if a part is missing, the model fills it in, so leave it out only when you do not mind what it picks. The camera presets in the generator will write the camera part for you; choose General if you would rather write it yourself.
One subject, one camera move
The single most reliable improvement is restraint. One subject, one camera move per clip. "A chai glass being poured, static camera, steam rising" comes back clean almost every time. "A chai stall at dawn with a vendor pouring, customers arriving, the camera orbiting then pushing in on the glass" asks for a short film and gets a mess.
If a story needs more than one shot, make more than one clip. Lumen Motion 3 and some others can hold 3 to 4 scenes in a 15-second clip when you describe them in sequence, but each scene still wants its own single subject and move. For anything longer than 30 seconds, make parts and join them with the free trim and converter tools.
12 prompts you can copy
Each row gives a model and length to try. The draft suggestion is for testing the wording cheaply; move to the final model once the clip looks right. All credits are as of October 2026.
| Use case | Prompt | Try on |
|---|---|---|
| Kirana D2C product | A glass jar of mango pickle on a wooden counter, label facing the camera. Slow push in. Window light from the left, shallow depth of field, warm tone. | Lumen Motion 3 Standard, 5 s, 9:16 |
| Flat walkthrough | A bright 3BHK living room with a balcony and modern furniture. The camera glides forward slowly toward the balcony doors. Soft morning daylight, no people. | Vela 2.5, 10 s, 16:9 |
| Wedding invite teaser | Marigold garlands and brass diyas on a red cloth. A gentle breeze moves the petals and the flames flicker. Slow tilt up. Warm golden evening light. | Lumen Motion 3 Standard, 8 s, 9:16 |
| Festival sale | A row of gift boxes with gold ribbon on a dark table, confetti falling slowly. The camera slides right along the boxes. Rich warm light, soft bokeh behind. | Draft on Orbit 3, final on Nova 2K, 5 s |
| Food | A plate of hot biryani, steam rising, a hand lifts one spoonful. Slow dolly in. Soft side light, shallow depth of field. | Lumen Motion 3 Standard, 5 s |
| Fitness | A woman doing one kettlebell swing in a gym. The camera circles slowly around her at waist height. Hard overhead light, slight haze. | Pulse Pro, 6 s, 9:16 |
| Travel | A houseboat moving slowly along a palm-lined backwater canal. The camera tracks alongside at water level. Hazy morning light. | Vela 2.5, 10 s, 16:9 |
| Real-estate exterior | A modern residential tower at dusk with lit balconies. The camera rises slowly from street level to the roof. Clear sky, warm window light, no text. | Nova 2K, 10 s, 16:9 |
| Saree fabric close-up | Extreme close-up of deep green silk with a gold zari border. The fabric ripples slowly as if lifted by hand. Static camera. Soft diffused light showing the weave. | Nova 2K, 5 s, 1:1 |
| Chai | A steel glass of chai being poured from a kettle, steam rising, slight slow motion. Static camera. Warm late-afternoon light from the side. | Nova Lite 6 s draft, Lumen Motion 3 Standard 5 s final |
| Monsoon street | A rain-soaked Mumbai street at night, people with umbrellas, shop lights reflected in puddles. Slow handheld walk forward. Neon and sodium light, shallow depth of field. | Director 4, 15 s, 16:9 |
| Temple morning | A stone temple gopuram at sunrise, pigeons taking off from the steps. Slow tilt up from the steps to the tower. Soft golden light, light mist. | Vela 2.5, 10 s, 9:16 |
Notice how every row has one subject, one camera instruction, one in-frame motion and one light note, and none of them mention a brand name, a logo or text on screen. Add the logo afterwards with the free video tools rather than asking the model to render it.
The dials that fix a prompt without rewriting it
When a clip is close, change a setting before you change the words.
The Avoid box is a negative prompt. If hands keep appearing in your product shot, type "hands, people" there. If the model adds captions, "text, watermark, logo". It is the quickest fix for the one thing that keeps spoiling an otherwise good clip.
Prompt strength sits at 0.5 by default. Raise it when the model is drifting from your wording and you want it followed more literally; lower it when the result feels stiff and you would like the model more room. Small moves, 0.1 at a time, are enough.
The Seed is optional and fixes the randomness. Leave it blank to get a fresh take each run. Once a run comes back good, note its seed and reuse it when you change only the length or resolution, so the final matches the draft you approved.
"Let the model expand my prompt" takes a short line and fleshes it out with setting and light. It helps when you have written five words. It gets in the way when you have written the shot precisely, so switch it off for the prompts in the table.
Draft cheap, then spend on the final
Nova Lite makes a 6-second draft for 48 credits, about ₹24, and Orbit 3 makes a 5-second one at 480p for 70. Use those to find out whether the words work. When they do, move to Lumen Motion 3 Standard at 115 credits for 5 seconds, or Nova 2K at 175 when you need the sharpest frame, or Vela 2.5 when the shot has to feel like cinema. You see the total on the Generate button before each run, and credits come back automatically if a model declines.
Turn sound off for Reels and Shorts where you will add a trending track; leave it on when the clip is going out on WhatsApp Business or a website on its own. Then finish on-device: resize for Reels, compress to under 16 MB for WhatsApp, and check the best format for Status and Reels before you post.
Frequently asked questions
How long should a text to video prompt be?
One to three sentences is usually enough: subject, camera, motion, light. The box takes 2,500 characters, but a longer prompt with several ideas tends to come back muddled.
Why does my clip look nothing like my prompt?
Most often the prompt describes a scene with no movement, or asks for two subjects and two camera moves. Cut it to one subject and one move, and raise prompt strength a little if the model is still drifting.
Should I let the model expand my prompt?
Yes when your prompt is short and you want the model to fill in setting and light. No when you have written the shot precisely and want it followed.
Which model should I test prompts on?
Orbit 3 or Nova Lite. A 6-second draft on Nova Lite costs 48 credits, about ₹24, so you can try three versions of a prompt for less than one final on a cinematic model.