Skip to content
AIupick

Guide for creators · 5 jobs

AI for creators: video, voice, music and images you may use

The models are good enough now. What separates usable work from AI slop is control: the same character and style from shot to shot, text that can be read, sound that fits the cut, and knowing what your plan lets you publish.

  1. 01

    Make a short video with the same character in every shot

    Generating each shot straight from the script gives you a slightly different person and place every time. Fix the look first, with reference images of the character, the setting and the style, and generate every shot from those references.

    Example

    A 30-second ad of a woman cycling through a city. One reference image of her, one of the street and one for the colour style; each of the eight shots calls them up, so her jacket is the same colour in the first shot and the last.

    How to do it

    1. 1Split the script into shots of a few seconds, each with one action and one camera move.
    2. 2Make or choose reference images for the character, the setting and the style.
    3. 3Generate the shots from the references: Ingredients to Video in Google Flow, or a saved element from the Element Library in Kling, called up with @ in the prompt.
    4. 4Edit the shots together, and add voice and music last, to the final cut.

    Built in

    Prompt to copy

    Turn this script into a shot list for a [length]-second video.
For each shot: duration (2 to 6 seconds), what the viewer sees, one action, one camera move, and which reference it uses (character, setting, style).
Describe the character and the setting in exactly the same words in every shot.

    Where it goes wrong

    • Hands, text and fast motion break most often. Keep them out of close-ups or cut around them.
    • A face that looks like a real person can get you into trouble. Use invented characters or people who agreed.
  2. 02

    Bring a still image to life

    Animations usually break when you ask for too much movement. Describe what the camera does and what moves in the scene separately, keep the movement small, and let the image do the rest.

    Example

    A photo of a lighthouse at dusk. "Camera: slow push in. Subject: the light turns, the waves move a little. Nothing new enters the frame." Four short versions, one keeper, extended once.

    How to do it

    1. 1Start from a sharp image with space around the subject.
    2. 2Write the prompt in two parts: what the camera does, and what moves in the scene.
    3. 3Use Animate in Midjourney with low motion, which also takes an uploaded image as the start frame, or Motion Control in Kling to copy the movement of a reference video.
    4. 4Make several short versions, pick one, and extend only that one.

    Built in

    Prompt to copy

    Camera: slow push in, eye level. Subject: [one small movement, such as hair moving in the wind]. Background: still. Mood: [mood]. Nothing new enters the frame.

    Where it goes wrong

    • Faces and hands distort during big movements. Keep the movement small near them.
    • Animating someone else's photo does not make it yours. Check the rights to the image first.
  3. 03

    Publish in other languages without re-recording

    You do not need to record again for each language. Dub the finished video and keep your own voice, then check names, numbers and terms in every language: that is where translation errors hide.

    Example

    A six-minute product video dubbed into Spanish and German. The checklist from the prompt has 14 items: product names, prices, units and one joke that does not translate. Your native-speaker check starts with that list instead of the whole transcript.

    How to do it

    1. 1Finish the edit in your own language first.
    2. 2Dub it with Dubbing in ElevenLabs or Translate video in Adobe Firefly, which keep the speaker's voice.
    3. 3Have a native speaker check names, numbers and product terms, or at least read the translated transcript.
    4. 4Clone a voice only with consent: your own, or that of someone who agreed in writing.

    Built in

    Prompt to copy

    Here is the transcript of my video. Target language: [language].
List every name, number, unit, product term and joke. For each, give the translation the dub should use and flag anything that could be misunderstood.

    Where it goes wrong

    • Translated speech often runs longer. Check the timing in the busiest scenes.
    • Some platforms ask you to label realistic AI voices. Check before you publish.
  4. 04

    Music that fits the cut, and that you may use

    Generate the music after the edit, to the cut's length and mood, not before. Get the stems so you can turn the instruments down under speech, and check what your plan lets you publish.

    Example

    A 90-second edit that turns from calm to upbeat at 0:40. The prompt asks for 90 seconds at 100 BPM with a build at 0:40 and a clean end. With the stems, you lower the drums under the voiceover between 0:10 and 0:30.

    How to do it

    1. 1Lock the edit and note its length and the moments the music should change.
    2. 2Describe mood, tempo and structure in the prompt. To fix one part, regenerate just that section with Edit in Udio, or rework single tracks in Suno Studio.
    3. 3Download the stems, with Get Stems in Suno, and lower the busy parts under the voice.
    4. 4Check the licence of the plan you were on when you made the track.

    Built in

    Prompt to copy

    Instrumental, [genre], [tempo] BPM, [length] seconds. Quiet 10-second intro, builds at [time], ends cleanly at [time] without a fade. No vocals. Mood: [mood].

    Where it goes wrong

    • What you may publish depends on the plan. Check its terms, and the tool page for what is confirmed.
    • A track that sounds like a famous song can still get a copyright claim on video platforms.
  5. 05

    Thumbnails with text people can read, in your own style

    Many image models still garble text inside images, and a feed of thumbnails in different styles looks like nobody is behind it. Use a model built for text, or add the text yourself, and save your style once.

    Example

    A video called "I tried five AI note-takers". The thumbnail says "Which one wins?" in quotes in the prompt, on the left, with you on the right, in the same colours as your last ten thumbnails.

    How to do it

    1. 1For text in the image, use Ideogram, which is built to render text, or leave space and add the text in an editor.
    2. 2Keep one style: Moodboards or Personalization in Midjourney, or Custom styles in Recraft.
    3. 3Look at the thumbnail at phone size, about the width of a thumb, before you publish.
    4. 4Keep two or three versions and test which gets more clicks, where your platform allows it.

    Built in

    Prompt to copy

    Thumbnail, 16:9. Large bold text "[3 to 5 words]" in the left third, high contrast. On the right: [subject], close up, looking at the text. Plain [colour] background. Nothing else in the frame.

    Where it goes wrong

    • Long text breaks. Keep it to five words or fewer.
    • A real person's face or a brand logo needs permission.

Rules for creators

  • Check commercial use and ownership on your plan before you publish or monetise anything.
  • Clone a voice only with the written consent of the person it belongs to.
  • Label synthetic media where the platform you publish on asks for it.
  • Do not prompt for other artists' names or characters you do not own.

Go deeper

Esc
↑ ↓ to moveEnter to open