Subscribe
AI Video
6 min read
axc
axc

How to Make a Short-Form AI Video, Step by Step

Eight steps from idea to posted short: set a vertical format, plan shots, make key images for consistent characters, generate takes in Runway, Kling or Google Flow, add sound and captions, and follow YouTube and TikTok rules for labeling AI content.

Olivetti-style poster reading Short-Form AI Video, How to make one step by step, with a vertical phone screen marked 9:16 on cream paper
Olivetti-style poster reading Short-Form AI Video, How to make one step by step, with a vertical phone screen marked 9:16 on cream paper

AI video tools generate short clips, which makes them a fit for short-form vertical video on YouTube Shorts and TikTok. But a finished short is more than one generated clip. It needs a hook, a few shots that match, sound, captions and, often, an AI label.

This guide walks through the whole process in eight steps. Each step points to the official help pages for the tools and platforms involved, so you can check the details yourself.

Before you start: what you need

  • An AI video tool that can make vertical clips. This guide uses examples from Runway, Kling and Google Flow. Our guide to the best AI video generators covers more options.
  • A way to add sound. Some video tools generate audio; a separate tool such as ElevenLabs can make sound effects.
  • An editor for captions and the final cut, such as Descript.
  • An account on the platform where you will post.
Diagram of the eight steps to make a short-form AI video: pick format, script, key image, generate, sound, edit, captions, label and post

Step 1: Pick the format and the platform rules

Decide where the video will go first, because that sets the shape.

YouTube's help page on uploading Shorts says a Short can be up to 3 minutes long and must have a square or vertical aspect ratio. A common vertical ratio is 9:16. Here is what the tools in this guide say:

  • Kling's text-to-video guide lists three aspect ratios: 16:9, 9:16 and 1:1.
  • Runway's Gen-4.5 help page lists 9:16 (720x1280) for image to video. As of October 2026, the same page lists only 16:9 for text to video. So with Gen-4.5, a vertical clip starts from a vertical image.
  • Google Flow's help page on creating videos says you choose the aspect ratio in your video preferences before you generate.

Set the aspect ratio before you generate. Cropping a wide clip to vertical afterward throws away much of the frame.

Step 2: Write a short script and a shot list

AI video tools generate short clips, so plan your video as a series of shots. Runway's Gen-4.5 page lists clip durations of 2 to 10 seconds. Kling's text-to-video guide advises keeping the visual content simple enough to complete within 5 to 10 seconds.

A simple structure for a short:

  1. Hook shot: the most striking image, first.
  2. Two to four middle shots: the story, the steps or the reveal.
  3. End shot: the payoff or a clear ending.

Write one line per shot: what is in it, what moves, and where the camera is. Those three things map directly onto how the vendors say prompts should be built. Kling's formula, for example, is subject, subject movement and scene, plus optional camera language, lighting and atmosphere.

If you are new to writing prompts, start with our beginner's guide to writing better AI prompts.

Step 3: Make a key image for each character or product

If the same person, pet or product appears in several shots, it has to look the same each time. The tools offer ways to help:

  • Google Flow uses "ingredients." Its help page says you can use the same images for "your character and key objects from one clip to the next." Google suggests references on a plain or segmented background.
  • Kling offers Elements. Its Elements guide says you upload 1 to 4 images, select the subjects, and describe their actions, which it says is useful for consistent looks across shots.
  • Runway image to video uses your image as the starting point. Its Gen-4 help page says the input image "acts as the first frame of your output video."

Make these images vertical, so every shot starts in the right shape.

Step 4: Generate each shot

Now turn each line of your shot list into a prompt. The vendors' guides agree on a few habits:

  • Describe motion as physical actions. Runway's prompting guide prefers "The woman smiles and waves" to abstract descriptions of a feeling.
  • For image to video, describe the motion, not the picture. Runway says repeating details already in the image "can lead to reduced motion or unexpected results."
  • Use one main camera move per shot. Kling's camera guide says asking for many moves at once "can create unstable results."
  • Phrase things positively. Runway says negative phrasing "may produce unpredictable or even opposite results."

Generate two or three versions of each shot and keep the best. Kling's image-to-video page notes that generation "isn't perfectly deterministic," so the same prompt can give a weaker result on a second try.

If a shot needs to land on a specific image, use start and end frames. Kling's start and end frames guide says the two images should be as similar as possible, because large differences may cause a shot switch.

Step 5: Add voice, music and sound

You have a few options for audio:

  • Generated audio. Google's video prompt guide for Veo and Gemini Omni Flash says to "clearly specify if you want audio" and to describe sound effects, ambient noise and dialogue in separate sentences.
  • Sound effects from text. ElevenLabs makes sound effects and ambient audio from a text prompt and says they are royalty free.
  • Voiceover. Descript lists AI speech with a voice clone or stock AI voices. Kling's lip sync guide can match a human character's lips to an uploaded voiceover or to its text-to-speech.

Note the platform rules here. YouTube's AI disclosure page lists AI-generated music that is the main focus of a video as something creators must disclose.

Step 6: Edit the shots together

Put the clips in order, trim the dead frames at the start and end of each, and cut to the beat of your audio.

  • In Google Flow, the Scenebuilder lets you arrange clips in a sequence, trim each one, preview the whole sequence and download the scene.
  • In Descript, the company says editing video "is as easy as using docs and slides."

Keep the hook in the first shot. If a middle shot does not add anything, cut it; short videos are short.

Step 7: Add captions

Captions let viewers follow the video without sound. Descript lists automatic transcription and says you can add subtitles in a single click. Check every word before you export, especially names and product terms.

YouTube's disclosure page lists caption creation with AI as a minor edit that does not need disclosure.

Step 8: Label AI content, check rights, and post

Each platform has its own rules for AI content.

YouTube. YouTube says it requires creators to "disclose AI-generated or altered content that seems realistic." Its list includes content that makes a real person appear to say or do something they didn't do, alters footage of a real event or place, or generates a realistic scene that didn't actually occur. You make the disclosure in YouTube Studio during upload, under "AI use" in the Attributes section. YouTube says creators don't need to disclose clearly unrealistic content, such as "someone riding a unicorn through a fantastical world," or minor edits. It also says that disclosing AI content alone "won't limit a video's audience or impact its eligibility to earn money."

TikTok. When TikTok introduced its AI-generated content label in 2023, it said its policy requires people to label AI-generated content that contains realistic images, audio or video. It said creators can use the label or other disclosures, such as a sticker or caption.

Rights. Check that your plan lets you use the output. Kling's credit guide says content from its free Basic plan is not for commercial use. YouTube's Shorts page also warns that copyright-protected material you did not clear could lead to a Content ID claim or removal.

Then post, and watch which hooks hold viewers. Use that to plan the next one.

A quick checklist

  • Vertical aspect ratio set before generating
  • One shot list line per clip
  • Key images for any repeated character or product
  • Two or three takes per shot
  • Sound added and cleared for use
  • Captions checked word by word
  • AI label set where the platform requires it
  • Plan terms allow your use

For a sense of what these tools cost per clip, see our AI video generator pricing comparison.

Sources

AI VideoTools

Written by

axc
axc

Edited by the llms.blog desk.

More in AI Video

All AI Video →

Popular this week