A good text to video prompt names five things in a fixed order: the camera work, the subject, the action, the setting, and the style and mood. Keep it to one scene and one camera move, put technical limits such as aspect ratio and length up front, and change one thing at a time when you revise. Below you get the formula, the vocabulary, ready-to-adapt examples and a repair routine.
In short
- Use one formula: camera, subject, action, context, style and mood. Google publishes the same five-part order in its Veo 3.1 prompting guide.
- One clip means one scene, one subject and one camera move. Clips are short, so a story needs several clips.
- Specific beats poetic. “Slow dolly-in, 35mm lens, soft window light” gives the engine something to act on. “Cinematic and beautiful” does not.
- Draft cheap, then upscale only the winner. At Take4K a 5‑second draft starts at 0.19 EUR.
- When a result misses, change one variable per attempt and keep a log of what worked.
What a text to video prompt actually does
A video model reads your sentence and decides what to render: who or what is in frame, how it moves, how the camera behaves and how the light feels. Anything you leave out, the engine fills in with its own defaults. Sometimes that is a pleasant surprise. More often it is a generic stock-footage look.
The job of a prompt is to remove guesses where it matters to you and leave freedom where it does not. A product shot needs strict control over the object, the angle and the background. A moody landscape can tolerate a looser brief.
Two habits separate prompts that work from prompts that burn credit:
- Describe what the camera sees, not what you feel. “A lonely, nostalgic scene” is a feeling. “A woman alone at a rainy bus stop at dusk, warm streetlight on wet pavement” is footage.
- Describe motion, not just a picture. A video prompt is not an image prompt with a longer name. If nothing in your sentence moves, the engine invents the movement.
The five-part formula
Google’s guide for its Veo model recommends a five-part structure: [Cinematography] + [Subject] + [Action] + [Context] + [Style & Ambiance]. The same order works well across most engines because it mirrors how a shot is planned on a real set.
| Part | What it answers | Example |
|---|---|---|
| Cinematography | How is it filmed? | Slow dolly-in, 35mm lens, shallow depth of field |
| Subject | Who or what is the focus? | A ceramic coffee cup with rising steam |
| Action | What happens? | Steam curls upward as a hand slides the cup across the table |
| Context | Where are we? | A sunlit kitchen counter, blurred plants behind |
| Style and ambiance | What is the look and mood? | Warm morning light, calm, clean commercial look |
Put together:
Slow dolly-in, 35mm lens, shallow depth of field. A ceramic coffee cup with rising steam on a sunlit kitchen counter, a hand slides it gently toward the camera. Blurred green plants behind. Warm morning light, calm, clean commercial look.
That is one sentence cluster and about 45 words. It is enough.
Put technical limits first
If your engine or workflow cares about aspect ratio, clip length or resolution, state them at the top or set them in the interface before you write the scene. A vertical 9:16 clip for Reels or Shorts is composed differently from a 16:9 clip for a website header, and it is cheaper to decide that before generating. If you shoot real footage too, the trade-offs are laid out well in this piece on vertical vs. horizontal video.
How long should a prompt be?
For most clips, 30 to 80 words is the sweet spot. Under 15 words, the engine guesses too much. Over 120, instructions start to compete and the engine drops some. If you need more detail, you usually need a second clip, not a longer prompt.
Camera, light, style and mood: the vocabulary
The cinematography slot is where beginners gain the most. Google’s guide calls it the most powerful tool for conveying tone and emotion, and lists moves such as dolly, tracking, crane, aerial, slow pan and POV shots. Use plain filmmaking terms. Engines have seen them captioned millions of times.
Camera movement
- Slow dolly-in: the camera glides toward the subject. Good for products and portraits.
- Pan or tilt: the camera turns left or right, or up or down, from a fixed position.
- Locked-off tripod: no movement. Use it when you want the subject’s motion to be the only motion.
- Handheld tracking: a slight shake, following the subject. Reads as documentary or UGC.
- Orbit: the camera circles the subject. Strong for hero shots of objects.
- Crane or aerial: a rising or high view that reveals scale.
Pick one. Asking for a dolly-in, an orbit and a whip pan in five seconds produces mush.
Lens and focus
A 35mm lens gives a natural, slightly wide look. An 85mm lens compresses the background and flatters faces. “Shallow depth of field, soft bokeh” isolates the subject. “Deep focus” keeps everything sharp, which suits architecture and landscapes.
Light
Light does more for perceived quality than any other word in the prompt. Try “soft window light,” “golden hour backlight,” “overcast flat light,” “neon signs reflecting on wet street,” or “hard midday sun with deep shadows.” Name the direction when it matters: side light, backlight, top light.
Style and mood
Name a look, not a vibe: “clean commercial,” “35mm film grain,” “muted documentary palette,” “high-contrast noir,” “bright pastel animation.” Then add one mood word, such as calm, tense or playful. Stacking six mood words cancels them out.
Prompt examples you can adapt
These are templates written by us. Swap the nouns for your own subject and keep the structure.
Product hero shot
Slow orbit, 50mm lens, shallow depth of field. A matte black wireless speaker on a polished concrete surface, small water droplets on the top. Soft side light from the left, dark gradient background. Premium, minimal, clean commercial look.
Why it works: one object, one move, one background, one lighting direction.
Social ad, vertical
Vertical 9:16, handheld tracking, 24mm lens. A young woman in a bright yellow raincoat jumps over a puddle on a city sidewalk and laughs toward the camera. Overcast daylight, saturated colors, energetic and playful.
Why it works: the handheld feel reads as authentic, which suits short-form social.
Restaurant or real estate b-roll
Slow tilt up, 35mm lens. A wood-fired pizza with melting mozzarella emerges from a brick oven, a chef’s hands slide it onto a wooden board. Warm firelight, glowing embers, rustic and inviting.
For real estate, swap in “slow dolly forward through an open-plan living room, floor-to-ceiling windows, late afternoon sun on oak floors, calm and spacious.”
Before and after
Generate two clips from the same description and change only the state: “a cluttered, dim garage” and then “the same garage, tidy, bright white light, shelves organized.” Keep the camera move identical in both so the cut feels deliberate.
Single emotional scene
Locked-off tripod, 85mm lens, shallow depth of field. An elderly man sits by a window reading a letter, his eyes slowly fill with tears. Soft overcast light, muted palette, quiet and tender.
Three-shot narrative
Short clips cannot hold a whole story, so build one from three prompts that share a character description and a look:
- Wide: “Locked-off wide shot. A cyclist in a red jacket rides along an empty coastal road at dawn. Cool blue light, light mist.”
- Medium: “Handheld tracking, medium shot. The same cyclist in a red jacket, pedaling hard, breath visible. Cool blue dawn light.”
- Close: “Slow dolly-in, close-up. The same cyclist’s face, smiling as the sun breaks over the sea. Warm gold light.”
Copy the character’s description word for word in every prompt. That is the cheapest way to keep a subject consistent. For longer sequences, a character consistency tool helps more than wording alone.
Text to video or image to video?
If a precise look matters, such as a logo, a product or a face, start from a still. Generate or upload the image, then animate it with a short motion prompt like “slow dolly-in, steam rising, soft flicker of candlelight.” You only describe the movement because the picture is already fixed. The image to video tool works this way.
Adapting prompts to different engines
Every engine has a personality. Some follow camera instructions to the letter and others favor their own cinematic taste. Some handle people better, some handle physics and fabric better. The same prompt can produce clips that look like they come from different studios.
Practical rules:
- Keep a master prompt in the five-part formula. Trim or reorder for each engine rather than rewriting from scratch.
- Run the same prompt through two or three engines at draft quality and compare. We published a side-by-side of one prompt across 13 engines to show how large the differences are.
- Note which engine suits which task. Many creators end up using one engine for people, another for product shots and another for stylized animation.
- Respect length limits. Most clips run around 5 seconds. Plan scenes that fit.
This is the reason we built Take4K around several engines in a single studio: you test a prompt on different engines without switching accounts or tools.
Fixing a prompt that misses
When the clip is wrong, resist the urge to rewrite everything. Diagnose first.
| Problem | Likely cause | Fix |
|---|---|---|
| Subject looks different each time | Vague description | Add fixed details: age, clothing color, hair, one distinctive item |
| Camera does nothing or does something odd | Missing or conflicting move | One named move, placed first |
| Generic stock look | No light or style specified | Add a light source and a look |
| Too much going on | Several actions in one clip | Cut to one action, split the rest into another clip |
| Wrong mood | Stacked or abstract mood words | One concrete look plus one mood word |
| Odd motion on hands or fabric | Complex action | Simplify the action, slow it down, try another engine |
A simple revision loop
- Generate a low-cost draft.
- Write down in one line what is wrong.
- Change one variable: the camera, the light or the action.
- Generate again and compare.
- When one draft is right, upscale only that one.
If you want to sharpen how you write revision instructions, the same logic applies to text, as in this guide on how to humanize AI text with instructions: say specifically what to change, not just that it feels off.
Common mistakes
- Writing a script instead of a shot. Dialogue, backstory and scene changes do not belong in a 5‑second prompt.
- Asking for several camera moves at once.
- Using negatives as the main control. “No people, no text” is weaker than describing an empty street.
- Changing five things between attempts. You never learn what helped.
- Skipping the aspect ratio. Recomposing a 16:9 clip into 9:16 crops out your subject.
- Upscaling every draft. Spend on resolution only after the motion is right.
Perspective
Most people think a bad AI video is a bad idea. In our experience it is usually a bad first draft, and the fix is cheap if you work in the right order.
We built Take4K around that belief. You see the exact price of each asset before you create it, you start with a low-resolution draft, and you upscale only the version that deserves it. That changes how you prompt. When a draft costs from 0.19 EUR, you can try three camera moves and two lighting setups and keep the best, instead of agonizing over a single expensive attempt.
The second belief is that no single engine is best at everything. Prompting well means matching a prompt to an engine, so we put several engines in one studio and let you compare them on the same description.
Our advice: write short, concrete prompts, change one thing at a time, and let the numbers on the screen guide how much you spend.
Vlad
Try these prompts in Take4K
Take4K turns text prompts or photos into video, with several engines in one studio and the price shown before you generate. Text to video starts from 0.19 EUR per 5‑second video. Image to video starts from the same price.
When a draft is right, the video upscaler takes it to Full HD or 4K, priced at 0.07 to 2.33 EUR per 5 seconds of video at 30 fps. If you want a character to speak, talking photo with lip sync runs 0.34 to 1.08 EUR per 5‑second video. Plans start at 11.99 EUR per month for Starter, or 132 EUR per year, and the pricing plans go up to Creator at 199.99 EUR per month.
FAQ
How long should a text to video prompt be?
Aim for roughly 30 to 80 words in a single scene description. Shorter prompts leave too much to the engine, and much longer ones tend to have instructions that conflict. If you need more detail, split the idea into two clips.
What is the best order for the parts of a video prompt?
Start with the camera, then the subject, the action, the setting, and finally style and mood. This is the same five-part order Google recommends in its Veo prompting guide. Technical limits like aspect ratio belong at the top or in the settings.
Can I reuse the same prompt on different engines?
Yes, and it is a good way to learn how each engine behaves. Keep one master prompt, run it as a low-cost draft on a few engines and compare. Then adjust the wording for the engine you choose.
How do I keep a character consistent across clips?
Repeat the same fixed description in every prompt: age, clothing, hair and one distinctive detail. Keep the lighting and style words the same too. Starting each clip from the same source image also helps.
Why does my video ignore part of my prompt?
Usually the prompt asks for too much in a 5‑second clip, or two instructions conflict. Cut to one subject, one action and one camera move. Then add detail back one element at a time.
