How to Make an AI Video From Text: Beginner Guide (2026)

To make an AI video from text, describe one short scene (who is in it, what they do, where it happens, how the camera moves and how it should look), then pick a model, an aspect ratio and a clip length and generate. A minute or two later you have a clip of 4 to 10 seconds. Longer videos are built like films: several clips, generated one shot at a time, then edited together.

This text to video AI tutorial covers the whole process: prompts the model can follow, model choice, fixing bad clips, sound, and building a 30 to 60-second video, with real costs per clip.

What text-to-video AI can do in 2026

A text-to-video model generates moving footage from a written description: you type a prompt and the model creates every frame. The newest models also create sound in the same pass. Google says Veo generates dialogue, sound effects and ambient noise along with the picture, while noting that natural speech, especially short speech, is still in active development (Google DeepMind).

No model turns one prompt into a long, finished film. Each generation is a single shot, and simple instructions work far better than complicated ones. So think in storyboards: one prompt, one shot.

The landscape also changes quickly. OpenAI closed its Sora app in April 2026 and shut down the Sora 2 video API in September 2026 (OpenAI), so Sora-based tutorials no longer apply. At the time of writing (October 2026), Google's Veo 3.1 and Kuaishou's Kling 3.0 are widely available through video apps, including ToonlyAI, so this guide uses them.

Text to video builds everything from your words, which makes it the fastest way to turn text into video with AI and test an idea. Image to video starts from a picture you supply, so you control the first frame, which matters for recurring characters, products and real photos. If you already have the image, read how to animate a photo with AI.

Before you start: aspect ratio, clip length and model

Make three decisions before you write the prompt:

• Aspect ratio. Use 9:16 for TikTok, Instagram Reels and YouTube Shorts, 16:9 for regular YouTube videos and websites, and 1:1 square (Kling 3.0 only) for feed posts. A vertical frame suits one centred subject with room above the head; our guide to AI videos for TikTok, Reels and Shorts covers vertical framing in depth.
• Clip length. Shorter clips give the model less time to drift. Use 4 to 5 seconds for one action or reaction, and 8 to 10 seconds for an action with a clear beginning and end, or a short line of dialogue.
• Model. Here's what the ToonlyAI Video Studio offers:

Step 1: Write your idea as a single scene

The most common beginner mistake is writing a topic instead of a scene. “A dragon story” is a topic. A scene has a specific subject doing one thing in a specific place, seen through a specific camera.

Google's Veo prompt guide lists the elements a video prompt can include: subject, action, scene, camera angle and movement, lens, visual style, lighting and mood, timing effects such as slow motion, and audio. Not all are required, but covering the first five leaves the model much less to guess.

Two more rules are worth memorising. Describe only what the viewer can see, not feelings or backstory. And use one main camera move per shot, tied to the subject: Kling's camera guide warns that a static camera combined with a push-in and an orbit sends mixed signals.

Put together, those rules give you this template:

[Style]. [Subject with two or three visual details] [does one clear action] [in a specific place, at a specific time]. [Shot type], [one camera move]. [Lighting and colour], [mood]. Sound: [one or two sounds].

Only have a rough idea? Type it as one line: the ToonlyAI studio can expand a short idea into a fuller cinematic prompt that you then edit. Before generating, delete anything that competes with your main idea, such as a second action or a conflicting camera move.

5 weak vs strong text-to-video prompts

Each weak prompt below names a topic. Each strong version describes one shot the model can actually film.

Vertical 9:16, polished 3D animation. A nervous lemon in a tiny trench coat tiptoes across a moonlit kitchen counter, then freezes as a grumpy orange in a security cap switches on a flashlight. Slow push-in on the lemon's guilty face. Warm under-cabinet light, playful suspense. Sound: soft footsteps, a flashlight click.

Anime style with soft cel shading. A girl on a rainy train platform at dusk lowers her yellow umbrella and smiles as sunlight breaks through the clouds. Medium shot, slow tilt up from her shoes to the sky. Pastel orange and teal palette, hopeful mood. Sound: light rain fading to silence.

Stop-motion claymation with visible fingerprints in the clay. A round grey cat in a tiny chef hat rolls out dough, sneezes, and vanishes in a puff of flour. Static eye-level medium shot. Warm kitchen light, cosy and funny. Sound: a small sneeze and a soft puff.

Cinematic live-action look. An old fisherman pushes a wooden boat into a misty lake at dawn and climbs aboard. Low-angle wide shot, slow dolly forward across the water. Shallow depth of field, cool blue mist with a warm sunrise glow, quiet mood. Sound: lapping water, distant birds.

Photoreal sci-fi. A small maintenance robot with one glowing blue eye rolls along the hull of a space station, stops, and looks up as a giant ringed planet rises behind it. Wide shot, slow crane up. Cold metallic light with an orange planet glow, a sense of awe. Sound: soft mechanical whirring, a low hum.

Step 2: Choose Veo 3.1 Lite, Veo 3.1 Fast or Kling 3.0

Start with Veo 3.1 Lite while you shape an idea: it's the cheapest model, so you can afford several versions. Once a prompt works, render the final on Veo 3.1 Fast for 1080p, or on Kling 3.0 for a 10-second shot, a square frame or smoother, more expressive motion. For a deeper comparison and a test you can run yourself, see Veo 3.1 vs Kling 3.0.

Step 3: Generate, review and change one variable

Generation usually takes one to three minutes. Watch each clip twice, checking that the action happens, the camera does what you asked, and faces, hands and the background hold together.

If something's wrong, change one thing at a time: the action, the camera move or the length. Rewrite everything at once and you can't tell which change fixed the problem.

Failed generations are refunded automatically. A clip that generates but misses the mark counts as a normal generation, which is why drafting on the cheapest model pays off.

Step 4: Decide whether to turn sound on

Veo 3.1 (Lite and Fast included) and Kling 3.0 can generate native audio: sound created together with the picture, so a door slam lands on the door. In the studio, sound is a toggle that changes the price. An 8-second Veo 3.1 Lite clip costs 50 stars with sound and 30 without.

Turn sound on when audio is part of the story: a line of dialogue, a crunch, a gasp. Describe it in separate sentences after the visuals, for example: Sound: rain on a tin roof, a distant dog barking.

For dialogue, Google's current Veo best practices advise against quotation marks, because quoted text can end up on screen. Use a colon instead: The lemon whispers: I didn't take the sugar. As a rule of thumb, about 20 spoken words fit in an 8-second clip.

Turn sound off if you'll add a trending sound, music or a voiceover later, or while drafting; it saves stars on every attempt. For dialogue syntax and fixes for garbled speech, see AI video with sound.

How to make a 30–60 second AI video from several clips

How long can AI videos be?

On ToonlyAI, one generation runs 4, 6 or 8 seconds on Veo 3.1 and 5 or 10 seconds on Kling 3.0. Some platforms go longer (Kling 3.0 reaches 15 seconds on Kling's own site), but every extra second is another chance for the model to drift. The reliable route to a 30 or 60-second video is several short shots edited together:

1. Storyboard. Write 4 to 6 beats: hook, setup, turn and payoff. Each beat becomes one shot.
2. Lock your character. Write one identity line per character (type, colours, clothing, one distinctive feature) and paste it unchanged into every prompt, as Google's best-practice guide recommends. For tighter control, start each shot from the same character image; see consistent characters in AI video.
3. Generate shot by shot with the same model, aspect ratio, style words and lighting, so the clips cut together cleanly.
4. Stitch the downloaded MP4s in any editor: your phone's built-in editor, CapCut, or the editors inside TikTok, Instagram and YouTube.
5. Caption and add sound so the story works muted, then layer music or a voiceover if your clips are silent.

Four 8-second clips make a 32-second story, a good fit when you want to make a 30-second AI video for Reels or Shorts:

Vertical 9:16, polished 3D animation with a miniature tilt-shift look. A tiny glossy red ladybug with seven black spots and big round cartoon eyes pushes a curled green leaf into a rain puddle. Extreme close-up at water level, slow push-in. Soft overcast light, curious mood. Sound: gentle rain, a small splash.

For shots 2 to 4, keep the style sentence and the ladybug description identical and change only the action and camera. AI fruit drama applies the same method to a full episode, clip by clip.

What an AI video clip actually costs

ToonlyAI uses pay-as-you-go stars, not a subscription: 200 stars cost $5, 500 cost $10 and 1,200 cost $20. Stars never expire, nothing auto-renews, and the studio shows the star cost before you generate.

The cheapest clip, a silent 4-second Veo 3.1 Lite video, costs 15 stars: roughly 25 to 38 cents depending on the pack. Here's how the four-shot ladybug story adds up:

• Silent drafts of all four 8-second shots on Veo 3.1 Lite: 4 × 30 = 120 stars.
• Final version on Veo 3.1 Lite with sound: 4 × 50 = 200 stars, exactly the $5 pack.
• Final version on Veo 3.1 Fast at 1080p with sound: 4 × 120 = 480 stars.
• Three 10-second Kling 3.0 shots with sound instead: 3 × 155 = 465 stars.

These are illustrative numbers, so budget for re-rolls: if two shots need a second attempt, the Lite project rises from 200 to 300 stars. Draft silently on Lite, then re-render only the winning prompts on a pricier model. Current packs are listed under pricing.