On this page
Chapter 1
What AI video generation is
An AI video model turns a description into a short moving clip. You don't film, light or animate anything: the model paints every frame from what it has learned about how scenes look and move. There are two main ways to drive it.
Text to video
You write a prompt (the subject, what happens, where, how the camera moves) and the model generates the clip from nothing. Ideal when you have no footage at all.
Image to video
You give the model a still as the first frame, such as a product photo or an approved storyboard frame, and it animates from there. The result stays much closer to the look you chose.
Each generation is a single short clip of a few seconds. A finished video is several clips joined together, with a voice, captions and often music on top: that assembly is the part most people underestimate.
Chapter 2
How video models differ
Every model makes trade-offs. Before choosing one, compare them on the things that change your result and your bill.
| What to compare | What it means | What to look for |
|---|---|---|
| Clip length | Most models render fixed or bounded lengths, typically from about 3 to 15 seconds per clip. | Plan scenes that fit the lengths the model renders, or you pay for seconds you trim. |
| Resolution | Common outputs are 720p and 1080p; premium models go up to 4K. | 720p is often enough for phones. Pay for 1080p or more when the video runs on big screens or in ads. |
| Aspect ratio | Not every model renders every shape: 9:16 vertical, 16:9 widescreen and 1:1 square. | Generate in the final shape. Cropping a widescreen clip to vertical loses most of the frame. |
| Motion and detail | Higher tiers usually handle complex motion, hands, fabric and lighting more convincingly. | Test the idea on a cheaper tier, then render the final on a better one. |
| Cost per second | Pricing is usually per generated second, and it rises steeply with quality: from a few cents a second for lite models to tens of cents for premium ones. | Multiply by every second you render, retries included. The storyboard step is how you keep retries cheap. |
| Audio | Some models can generate sound, but speech in Arabic dialects is not their strength. | Generate the pictures silent and add a real voice-over and music on top. |
Vocla's quality tiers
- Lite150 credits/sFast and economical, 4, 6 or 8-second clips. Ideal for social clips and for testing an idea. On every plan with video.
- Standard400 credits/sRicher motion and detail, flexible 3 to 15-second clips, every format. Professional and up.
- Pro1,000 credits/sThe highest detail, for campaigns and showcase work: 1080p on Business, up to 4K on Enterprise.
Vocla routes each tier to a current AI video model and may change the model behind a tier as better ones appear; the tier, its credit rate and your plan limits stay the same.
Chapter 3
Why the voice and captions matter more than the visuals
In short-form video, people forgive a simple picture. They don't forgive a voice that sounds wrong, a dialect that isn't theirs or words they can't read on mute. That's why Vocla builds a video around the voice, not the other way around.
The voice carries the message
The words, the tone and the accent decide whether the viewer feels addressed. A Saudi voice for Riyadh or an Egyptian one for Cairo does more than a prettier shot.
Captions do the work on mute
Plenty of people watch with the sound off. Burned-in captions, in a style that's easy to follow, keep the message intact for them.
Timing comes from speech
Scenes should change with the sentences. Vocla voices the script first and times every scene to the speech, so the picture follows what is being said.
Text inside frames is the weak spot
Video models garble lettering, Arabic above all. Keep every word out of the picture and on a separate caption track that renders Arabic correctly, right to left.
Chapter 4
How to write scene prompts that look good
A good scene prompt reads like a director's note: who or what is on screen, what happens, where, how the camera behaves, how it's lit and what it should feel like. Write it in English, which video models follow best, even when your script is in Arabic.
The six parts of a scene prompt
- 1. Subjecta brass dallah and a small cup of Arabic coffee
- 2. Actioncoffee being poured, steam rising
- 3. Settingon a wooden table by a window
- 4. Cameraclose-up, slow push-in
- 5. Lightingwarm morning light, shallow depth of field
- 6. Stylecinematic, rich golden tones
Close-up of Arabic coffee being poured from a brass dallah into a small cup, steam rising, on a wooden table by a window, slow push-in, warm morning light, shallow depth of field, cinematic, rich golden tones
Habits that help
- One subject and one action per scene: a phone screen is small.
- Name the camera move. Slow push-ins, tilts and top-down shots suit vertical video.
- Repeat the same style words at the end of every prompt so the scenes feel like one video.
- Describe what you want, not what you don't: models often ignore "no".
What to leave out
- On-screen textTitles, prices, signs and Arabic words come out garbled. Put words in the captions instead.
- Logos and brand namesModels invent fake logos or distort real ones. Show unbranded objects and add your brand in the voice-over.
- Real or famous peopleDon't ask for a real person's likeness. Use a generic person, hands, or a figure seen from behind.
- Crowded scenesMany subjects and actions at once turn into visual noise. Split them across scenes.
- Contradictions"Bright sunny night" or "static drone fly-through" confuse the model. Pick one light and one move.
Chapter 5
Copy-paste prompt library
Twelve scene prompts that follow the six-part recipe. Copy one, swap the subject for yours and keep the camera, light and style words.
- FoodArabic coffee pour
Prompt
Close-up of Arabic coffee being poured from a brass dallah into a small handleless cup, steam rising, warm morning window light, shallow depth of field, slow push-in, cinematic, rich golden tones
One subject, one action and a gentle move: a calm, premium opening.
- RestaurantsCharcoal grill at night
Prompt
Close-up of skewers of grilled meat and vegetables over glowing charcoal, flames flickering, smoke curling upward, night market atmosphere, warm firelight, slow motion
Fire, smoke and slow motion give texture without needing any text.
- Real estateLiving room reveal
Prompt
Slow dolly forward into a bright modern living room with floor-to-ceiling windows overlooking a city skyline at golden hour, minimalist furniture, soft natural light, wide lens, calm and airy, photorealistic
A forward move reads as "walking in", which is what a tour should feel like.
- E-commercePerfume product hero
Prompt
A glass perfume bottle without a label on a dark marble pedestal, slow orbit around the bottle, soft rim light, fine mist drifting in the air, black background, luxury commercial style, sharp focus
An unlabelled bottle avoids garbled branding; your brand lives in the voice.
- BeautySkincare texture
Prompt
Macro shot of a creamy skincare texture swirling slowly on a pastel pink surface, soft diffused light, glossy highlights, clean and fresh aesthetic
Macro textures look rich in a tall frame and never need a face.
- TravelDesert dunes at sunset
Prompt
Aerial drone shot gliding over rippled sand dunes at sunset, long shadows, warm orange and pink sky, a single tyre track across the sand, epic and peaceful, cinematic color grade
Landscape, light and one guiding line: an easy, striking establishing shot.
- HospitalityOld-town alley at blue hour
Prompt
Slow walk-through of a narrow old-town alley with stone walls and hanging lanterns at blue hour, warm light spilling from doorways, gentle handheld camera, nostalgic, cinematic
The handheld feel makes it personal, like a creator walking you through.
- FitnessRooftop stretch at sunrise
Prompt
A person in plain sports clothes stretching on a yoga mat on a rooftop at sunrise, seen from behind, city in the soft distance, gentle handheld feel, warm light, calm and motivating
Seen from behind keeps the person generic and the mood front and center.
- EducationBuilding blocks
Prompt
Top-down view of colorful wooden blocks being stacked into a tower on a light desk, soft daylight, playful and clear, stop-motion feel
A visual metaphor for "step by step" with no writing for the model to garble.
- TechGlowing phone
Prompt
Close-up of a smartphone lying on a dark desk, its blank screen emitting a soft blue glow that lights the surroundings, subtle floating particles, slow push-in, futuristic, moody lighting
A blank, glowing screen suggests an app without fake interface text.
- AutomotiveCoastal drive
Prompt
Tracking shot of an unbranded white SUV driving along a coastal highway beside a calm turquoise sea, bright midday sun, smooth gimbal movement, clean commercial look
"Unbranded" heads off invented badges; the tracking move sells the motion.
- Any brandAbstract background loop
Prompt
Slow-moving abstract waves of teal and purple light flowing across a dark background, soft glow, smooth seamless motion, calm and premium
A safe backdrop for a statement, an offer or a quote spoken by the voice.
Chapter 6
Work storyboard first
Rendering video is the expensive step, so make every decision before it. A storyboard of stills, one per scene, shows the look of the whole video for a fraction of the cost of rendering it.
- 01ScriptHook first, one sentence per idea
- 02VoiceGenerated first, in the dialect you chose
- 03Scene plan3–8 s scenes timed to the speech
- 04StillsOne frame per scene to approve
- 05RenderOnly the scenes you approved
- 06MixVoice, captions and music on top
- Fix the hook frame first: if the first still doesn't stop you scrolling, the video won't either.
- Check consistency: the same light, palette and style across every still.
- Change one prompt and regenerate that still, instead of redoing everything.
- Approved stills become the first frame of each clip, so the video stays close to what you saw.
Chapter 7
Step by step: generate a video with Vocla
The whole flow, from a blank page to a downloaded MP4.
- 1
Write and time the script
Write a hook you can say in about two seconds, then one short sentence per idea. Check the length with the TikTok script planner.
- 2
Choose the voice and the finish
Open Video in Vocla, paste the script, then pick a voice and dialect, the format (9:16, 16:9 or 1:1), a quality tier, a caption style and whether to add music.
- 3
Let Vocla voice and plan
Vocla generates the voice-over, splits it into scenes of three to eight seconds timed to the speech and writes an English visual prompt for each.
- 4
Edit the scene prompts
Read every prompt against the six-part recipe. Make the subject specific, name the camera move and keep the style words consistent.
- 5
Approve the storyboard
Generate a still per scene. Rework any prompt that misses and regenerate that still until each scene looks right.
- 6
Render, review and download
Check the quote, render, then watch the video on your phone. Regenerate a single scene if needed and download the MP4 with your voice, captions and music.
Chapter 8
Plan the cost before you render
Vocla charges video in credits per rendered second, and the rate depends on the quality tier. Each scene is rounded up to a clip length the tier's model renders, so scene lengths matter as much as the total.
| Tier | Credits per rendered second | Clip lengths | Plans |
|---|---|---|---|
| Lite | 150 credits/s | 4, 6 or 8 s | Every plan with video |
| Standard | 400 credits/s | 3 to 15 s | Professional and up |
| Pro | 1,000 credits/s | 4, 6 or 8 s | Business and Enterprise |
Example: a 30-second Reel
Five scenes of six seconds each, voiced in Saudi Arabic, with storyboard stills and a music bed. A 30-second video needs Professional (Starter videos go up to 15 seconds).
- Scenes on Lite5 × 6 s × 150
- 4,500 credits
- Scenes on Standard5 × 6 s × 400
- 12,000 credits
- Storyboard stills5 × 200
- 1,000 credits
- Music bed (default)
- 120 credits
- Voice-over
- By the speech model
Why scene lengths matter
On Lite, a 15-second Starter video planned as three 5-second scenes renders three 6-second clips: 18 rendered seconds, or 2,700 credits. Planned as 4 + 5 + 6 seconds, it renders 4 + 6 + 6: 2,400 credits. Your monthly allowance counts the video's own length either way.
These are estimates from the default rates at the time of writing. Rates can change, and the quote shown in the app before you render is the exact figure. Failed scenes are refunded.
Chapter 9
Common mistakes (and the fix)
Most disappointing AI videos fail for the same few reasons. None of them is about the model.
Starting from the pictures
Write and voice the script first. Scenes timed to a finished voice-over feel intentional; pictures with a voice added later feel random.
Asking the model for text
Prices, titles and Arabic words come out garbled. Leave every word to the caption track.
Logos and real people
Ask for unbranded objects and generic people, and say the brand name in the voice-over.
A new style every scene
Repeat one style tail, the same lighting and palette words, at the end of every prompt.
Too much in one shot
One subject and one action per scene. If you need two ideas, you need two scenes.
Rendering before the stills
Approve the storyboard first; changing a still costs a fraction of re-rendering a clip.
A script that doesn't fit
Too many words for the length means a rushed voice. Time the script, then cut words instead of speeding up.
Forgetting the frame edges
Platforms draw buttons over the sides and bottom of vertical video. Keep subjects and captions clear of them.
Chapter 10
Questions about AI video generation
What is AI video generation?
Using an AI model to create video clips from a written description (text to video) or from a starting image (image to video), instead of filming. Tools like Vocla Video add the parts a model doesn't do well: the voice-over, captions, music and joining scenes into one video.
What's the difference between text to video and image to video?
Text to video generates a clip from a prompt alone. Image to video starts from a still you provide, such as a product photo or an approved storyboard frame, so the result stays closer to a look you've chosen.
Can AI video models write Arabic text in the video?
Not reliably: lettering usually comes out garbled, Arabic especially. Vocla never asks the model for text and burns captions in from a separate track that renders Arabic correctly.
How long can an AI-generated video be?
Each clip is short, usually a few seconds. Longer videos are built from several scenes. In Vocla, videos go up to 15 seconds on Starter, 60 seconds on Professional, 120 seconds on Business and 180 seconds on Enterprise.
How much does it cost to generate a video with AI?
It depends on the length and the quality tier. In Vocla, video is charged in credits per rendered second (by default 150 on Lite and 400 on Standard), plus the voice-over and optional storyboard stills and music. You always see the quote before rendering.
Which language should I write prompts in?
English: video models follow English prompts most reliably. Your script and voice can be in any Arabic dialect; Vocla writes the scene prompts in English for you, and you can edit them.
Can I use AI-generated videos commercially?
On Vocla, commercial use is included on paid plans, and every plan with video is a paid plan. You remain responsible for your script and the claims in your ads.