How to Make AI Videos for TikTok, Reels & Shorts (2026)

Here's how to make AI videos for TikTok, Instagram Reels and YouTube Shorts in one sentence: generate in vertical 9:16 from the very first frame, write your hook into the opening second or two, and stitch three to five short AI clips into a 20–45 second post with captions and sound. One vertical master can then go to all three apps.

Below you'll find a spec cheat sheet, five hook formulas, a clip-stitching workflow and a weekly plan for a faceless channel. New to prompting? Start with how to make an AI video from text.

Spec cheat sheet: TikTok vs Reels vs Shorts

Specs change often, so treat this as an October 2026 snapshot. Where a platform doesn't publish a clear figure for organic posts, the table says so.

Sources: YouTube's Shorts help page, Instagram's Reels help page and TikTok's in-feed ad specs, which accept .mp4 and .mov files up to 500 MB. TikTok's numbers come from its ad documentation, so treat them as a guide for organic posts, not a hard rule.

Plan vertical from the first frame

Choose 9:16 in the generator instead of cropping a landscape clip later. Cropping 16:9 down to 9:16 keeps less than a third of the original width, which often means losing the second character, the hands or the punchline. In ToonlyAI, all three video models (Veo 3.1 Lite, Veo 3.1 Fast and Kling 3.0) offer 9:16, and Kling 3.0 also offers 1:1.

Then compose for a tall, narrow frame:

• One subject, centred or slightly high. It reads instantly on a phone and leaves room for captions below.
• Move toward the subject or along the vertical. Push-ins, tilts, crane moves and slow orbits suit 9:16; wide sideways tracking shots tend to lose the subject.
• Stack depth, not width. Place a foreground prop, the character and a background detail one behind the other.
• Overlap two characters. Use an over-the-shoulder shot, or put one closer to the camera.
• Keep the edges quiet. App buttons and captions sit at the top, bottom and right, so keep the key action away from them.

Starting from a picture? Crop it to 9:16 before you animate it; how to animate a photo with AI covers photo prep. Our AI video prompts guide lists more camera moves that work in vertical.

Which model? Veo 3.1 Lite (720p) is the affordable option for drafts and simple scenes, Veo 3.1 Fast adds 1080p for a sharper final, and Kling 3.0 is built for smooth, expressive motion. Veo 3.1 vs Kling 3.0 compares them in detail.

The first 2 seconds: 5 hook formulas for AI clips

TikTok's creative guidance for advertisers says to grab attention in the first three seconds and skip long intros (TikTok video advertising guide). That guide is a few years old and written for ads, so treat it as directional. With AI video, the hook isn't something you find in the edit: you write it into the prompt.

Each formula below has a copy-paste 9:16 prompt. Swap in your own characters and setting.

1. Start mid-action

Skip the setup. The clip opens with the event already happening (a chase, a fall, a spill) and the viewer stays to see how it ends.

Vertical 9:16. A tiny clay snail sprints straight toward the camera across a kitchen counter, shell wobbling, as a huge drop of honey falls behind it in slow motion. Low-angle close-up, the camera slowly pulls back to keep pace. Bright claymation style, warm morning light. Sound: squelchy footsteps and a rising whoosh as the drop falls.

2. The impossible image

Open on something that can't exist: a world inside a teacup, a whale in a swimming pool, a city made of bread. Scale tricks and odd materials are cheap to generate and hard to scroll past.

Vertical 9:16. Extreme close-up of a teacup on a sunny windowsill; inside it, a miniature ocean where a tiny sailboat rides a wave. Slow push-in toward the boat. Miniature photography look, shallow depth of field, soft golden light. Quiet ambient sound of waves and distant seagulls.

3. A line straight to camera

A character looks into the lens and says one short line that raises a question. Write it after a colon, without quotation marks: Google's Veo guidance warns that quoted speech can appear as on-screen text (Veo best practices). For more on dialogue, see AI video with sound.

Vertical 9:16. A grumpy lemon detective in a tiny trench coat leans toward the camera inside a dim fridge, one eyebrow raised. Medium close-up, static camera. Polished 3D animation, cool blue fridge light. The lemon says in a low, gravelly voice: Somebody in this fridge is lying. A faint fridge hum in the background.

4. The reveal pull-back

Start tight on a detail, then pull back to reveal context that changes its meaning. The first frame asks the question and the pull-back answers it.

Vertical 9:16. Extreme close-up of a cat's wide, nervous eyes. The camera slowly pulls back to reveal the cat sitting in the pilot's seat of a small propeller plane above the clouds, paws on the controls. Polished 3D cartoon style, bright daylight. Steady engine hum, then one surprised meow.

5. Something is off

Show a normal scene with one thing wrong: a book floating, a shadow moving on its own. Viewers keep watching to resolve it, which suits spooky and mystery series.

Vertical 9:16. A quiet library at night, tall shelves lit by a single desk lamp. In the background, one book slides out of a shelf on its own and hangs in the air. Slow dolly-in toward the floating book. Storybook illustration style, deep blues and warm amber light. Ambient sound of a ticking clock and a soft page rustle.

From one 8-second clip to a 30-second post

ToonlyAI clips run 4 to 10 seconds: Veo 3.1 Lite and Fast make 4, 6 or 8-second clips, and Kling 3.0 makes 5 or 10-second clips. A full story needs several, which you can join in any free editor, such as CapCut or the tools built into TikTok, Instagram and YouTube.

1. Write a beat sheet. Three to five beats (hook, setup, twist, payoff, optional loop-back). One beat becomes one clip.
2. Lock the look. Reuse the same character description, style words and lighting in every prompt. See how to keep characters consistent.
3. Draft cheaply. Test uncertain beats as silent 4-second Veo 3.1 Lite clips (15 stars each).
4. Generate the finals at full length, with sound where you need dialogue or effects. Each clip usually takes one to three minutes and downloads as an MP4.
5. Assemble. Import the clips in order, trim slow frames at each end, and cut on movement so the joins feel intentional.
6. Finish. Add captions, music or voiceover and your text hook, then export one vertical master. Mixing 720p and 1080p clips is fine inside a single vertical project.

What does a full post cost? Example builds, one take per clip:

Stars come in packs (200 for $5, 500 for $10, 1,200 for $20), so a 200-star post costs roughly $3.33–$5. Budget extra for redos: failed generations are refunded automatically, but a clip that simply isn't what you wanted still costs stars. See current pricing.

Make it loop

A loop invites a replay. Plan the last beat so its framing and pose echo the first, or end on a line that sets up the opening, then cut on a frame that flows back into frame one.

Series formats that work for AI shorts

Recurring formats are faster to produce and give viewers a reason to follow. A few that suit AI clips:

• Episodic character drama. The same cast in a new situation each episode, ending on a cliffhanger. Fruit drama is 2026's best-known example; see how to make AI fruit drama videos.
• Talking objects. An everyday object with a face and strong opinions, like a mug that dreads Mondays. See AI talking object videos.
• Tiny adventures. Miniature characters in giant everyday places; the scale is the hook.
• Midnight stories. Short spooky scenes built around one unexplained moment.
• Mascot explainers. A recurring character explains one fact or tip per episode.

Each maps to a category in ToonlyAI's story starters, which pair a ready-made prompt with a concept image you can adapt. Whatever the format, change something meaningful every episode: YouTube doesn't monetize content that looks templated or mass-produced (see monetizing AI videos on YouTube).

Captions, sound and on-screen text

ToonlyAI's sound toggle decides whether the model generates audio. Turn it on when dialogue, effects or ambience must match the picture; turn it off when you'll add a trending sound or voiceover in the app. Silent clips cost less: an 8-second Veo 3.1 Lite clip is 30 stars silent versus 50 with sound.

Adding copyrighted music on YouTube? Keep the Short under a minute. According to YouTube's Shorts help page, Shorts longer than one minute with an active Content ID claim are blocked globally.

• Caption every spoken line so the story works for anyone watching muted.
• Keep text in the centre band, away from the bottom edge and right-hand side where usernames and buttons sit.
• Use TikTok's safe-zone files. The safe area changes with orientation and caption length. We couldn't find official numbers from YouTube or Instagram, so check each app's preview.
• Keep the text hook short: a few words, readable at a glance.
• Spell names the same way every episode and time captions to the speech.

Posting, AI labels and cross-posting

Upload the original export from your editor to each app, not a copy downloaded from another platform, which may carry that app's branding and an extra round of compression. Write each caption for its audience. On YouTube, make sure the file is vertical or square and three minutes or shorter, or it won't count as a Short.

On labels: all three platforms ask you to label realistic AI content, and some label automatically when a file carries C2PA Content Credentials. Clearly animated content is treated differently; YouTube, for example, doesn't require disclosure for clearly unrealistic or animated videos. Policies change, so the details live in one place: Can you monetize AI videos on YouTube? Rules and labels.

A weekly batch workflow for a faceless AI channel

A faceless channel, where you never appear on camera, runs best on a fixed weekly rhythm. Here's a simple version for five posts a week:

1. Monday: plan. Pick one series, write five beat sheets of three to four beats, and update your character sheet.
2. Tuesday: draft. Test the riskiest beats as cheap silent drafts and fix the prompts.
3. Wednesday: generate. Produce the final clips, with sound where needed, working in batches.
4. Thursday: edit. Assemble, caption, add sound, pick cover frames and export five vertical masters.
5. Friday: post and review. Schedule across the three apps, then check each platform's retention data to see which hooks held viewers.

Example budget: five posts × four 8-second Veo 3.1 Lite clips with sound = 20 clips × 50 stars = 1,000 stars. Add a buffer for redos and one 1,200-star pack ($20) covers the week. Stars never expire, so leftovers carry over.

Planning a full week at once also makes it obvious when two episodes are too alike, which is exactly the repetition you want to avoid.