You type "a two-minute video introducing my nonprofit, with smiling volunteers" into a video generator. You get a few seconds of floaty footage, faces that change halfway through, a hand with six fingers. Nothing you can use.
It's not that the tool is bad: you asked it for an entire film in one sentence. The right method is the one filmmakers use: shot by shot.
The main families of tools
"AI video" covers several families of tools. You'll often combine two or three of them.
- Text to video: you describe a scene and the tool generates a short clip. Handy for exploring, hard to control.
- Image to video: you provide a still image and the tool animates it. This is the most reliable family for keeping a consistent style and character.
- AI-assisted editing: editing software that cuts out silences, suggests transitions, reframes for vertical formats or finds a specific moment in your raw footage.
- Avatars and voices: a virtual presenter who reads your text, or a synthetic voice for narration.
- Automatic captions: the audio track is transcribed, then turned into formatted captions.
Why you don't make a film from a single prompt
A video generator produces short clips. The longer the clip and the busier the scene, the more likely the model is to drift: the character's face changes, the set transforms, objects appear and disappear.
One shot, one idea, a few seconds. Length is something you build in the edit.
A one-minute video is often a dozen short shots put end to end. Each one is simple, so each one has a good chance of working.
The method in eight steps
1. The idea in one sentence
If you can't sum up your video in one sentence, it isn't ready. "A baker opens her shop at dawn and serves her first customer." Anything that doesn't serve that sentence gets cut from the project.
2. A short script
Write what we see and what we hear, in order. An assistant like ChatGPT, Claude or Le Chat can help:
You are a short-form video scriptwriter. Based on this idea: [your idea], write a 30-second script. For each moment, say what we see and what we hear (voiceover, sound, music). No more than 60 words of voiceover.
3. Breaking it into shots
Turn the script into a list of shots, each a few seconds long. For each shot: a number, what happens, the target length. If a shot contains two actions, split it in two.
4. A reference image
This is the step that saves your consistency. First generate a still image of your character and setting with an image generator, then reuse it as the starting point (image to video) for every shot. To get that image right, the five-layer grid and our guide to generating an image with AI walk you through it.
5. The shot prompt
A video prompt looks like an image prompt, plus movement. Five elements:
- Subject: who or what, as seen in the reference image.
- Action: just one, and simple. "She flips the sign on the door."
- Camera movement: static shot, slow push-in, pan from left to right, handheld camera.
- Duration: a few seconds, staying within what the tool offers.
- Mood: light, weather, pace. "Bluish dawn light, steam on the windows, slow and calm movement."
Put together:
The baker from the reference image slowly flips the "Open" sign on the glass door of her shop. Medium shot, static camera. Bluish dawn light, light steam on the glass, calm atmosphere. Duration: a few seconds.
Describe visual facts, not emotions. You can't film "a warm atmosphere," but "amber light and steam on the glass" films beautifully.
6. Several attempts per shot
Generation is random: the same prompt gives different results. Run several attempts for each shot and keep the best one. When a shot doesn't work, change just one element at a time (often the camera movement or the action) and note what it changed.
7. Editing
Assemble the shots you kept in script order. Trim the beginnings and ends of clips, which are often messier. A failed shot can sometimes be saved by keeping just one second of it.
8. Sound and captions
Sound is half the experience: voiceover, sound effects, music, then automatic captions. Always proofread the captions, since names and numbers are often transcribed wrong. Lots of people watch with the sound off, so captions aren't optional.
Pitfalls to know about
- Character consistency. From one shot to the next, the hair, clothes or face change. Fix: same reference image, same description word for word in every prompt.
- Weird movements. Hands merging, legs sliding, objects warping. Fix: simple actions, slow movements, shorter shots.
- On-screen text. Generators are bad at writing. Add titles and signs in the edit, not in the prompt.
- Real people's voices and faces. Never reproduce someone's voice or face without their explicit consent, even as a joke between friends.
- Consent. Before animating a photo of someone close to you, ask them. For a child, ask the parents.
- Music and rights. A well-known song added to your video can get it blocked or taken down. Use royalty-free music or music generated with a tool whose terms allow your use.
The checklist before publishing
- The idea still fits in one sentence.
- The character is recognizable from one shot to the next.
- No warped hands, faces or objects are visible.
- On-screen text was added in the edit and proofread.
- Captions are proofread, names included.
- No real person appears or speaks without their consent.
- You have the right to use the music and sounds.
- The tools' terms allow your use, especially commercial use.
- The video is labeled as AI-generated if someone could mistake it for real.
Going further
On Ovrelia, the Prism island helps you understand where generated images come from, why two attempts never look alike, and how to go from intention to visual facts. The first chapter, "From Noise to Image," is free and plays in your browser, no account needed: play it now. And to test your AI knowledge, launch the friend challenge: ten timed questions and a score to beat.
Frequently asked questions
Can you make a video with AI without knowing how to edit?
Yes, AI-assisted editing tools handle much of the technical work. What you mostly need is a clear script and a breakdown into short shots; then putting it together becomes simple.
Why does my character change from one shot to the next?
Each shot is generated independently, so the model reinvents the details every time. Start from the same reference image and repeat the same character description in every prompt.
Should I generate the whole video in a single prompt?
No. Generators produce short clips and drift over time. Break your video into shots a few seconds long, generate them separately, then assemble them in the edit.
Am I allowed to use someone's voice or face?
Only with their explicit consent. Reproducing a real person's voice or face without consent can deceive people, harm that person and expose you legally.