How to make an AI commercial

What is an AI commercial, and how is this guide different?

An AI commercial is a scripted, cinematic ad in which AI generates the actors, sets and camera work. It looks like a produced spot or brand film rather than a phone video. What AI commercials are, why brands use them and who makes them is covered in AI commercials: what they are and how brands get them made. This guide covers only how to make one.

Public examples of the style are on AI brand film ads. They come from posts on X and were not made on or for Anglr.

Tools creators use for it

The counts show how many of the reviewed commercial and brand film tutorials used each tool.

Based on 21 public tutorials posted between 11 May 2026 and 12 Oct 2026, reviewed by Anglr.

Should you use one prompt or a full pipeline?

Use a full pipeline for anything you will deliver to a brand; use one prompt to test an idea. The sources split clearly on this.

Seven authors publish single-prompt spots: one block of text describing the presenter, setting, light, camera, product shots and sound (@I_Muhammadali44, @commonman_16, @Itswsm105f, @Lianaalane, @aiwithaayat, @alixbuilds, @Hope_Ai01). These are quick and show what a model can do in 15 seconds.

The longer tutorials build the commercial in stages instead, because a single prompt gives little control over which actor, which product angle and which shot order you get. The steps below follow that route.

How do you make an AI commercial step by step?

You make it in seven steps. This is our synthesis of the tutorials, with the points where they agree and where they differ.

1. Write a short script with one idea

Write the script as scenes: what we see, what we hear, how long. Keep it to one idea a viewer can repeat. @JuliaClarky frames an everyday food product as a heist film; @lr_mvcreative builds a spot around one visual rule broken at one moment. If you work from a brief, the script and angle may already be given; the craft is in staging it.

2. Draw the whole spot as a board before any video

Turn the script into a storyboard or production board and check it before spending video credits. This is the step most sources share. Estefania Freites builds a 15-frame board in a 5x3 grid with camera angle, sound and music direction per frame, then checks continuity and shot order. @MonetizationDon storyboards in ChatGPT first. Alber Adilov makes keyframes in Soul Cinema or GPT Image 2. Janvi Shah generates scene images and keeps only the frames that fit. Aneke Victor Ebuka plans actions, reactions and transitions for the whole sequence first.

Alber Adilov notes one practical split: use GPT Image when a frame needs readable text, and Soul Cinema for characters.

3. Build a locked sheet for each asset

Make one reference sheet each for the character, the product and the location, and reuse them in every scene. Abhinav Kalyan makes a two-panel character sheet (full body and face close-up) and a front, back and top product sheet so the model does not invent the sides it cannot see. Estefania Freites makes a product sheet at front and three-quarter, a character sheet on plain grey and a location sheet at a three-quarter angle. Aneke Victor Ebuka fixes face, hair, outfit and proportions per lead, and treats the location the same way. Ian Ramírez Rivera reports that Seedance 2.5 treated location, wardrobe, character and light references as hard constraints.

4. Have an LLM write the scene prompts from the assets

Give Claude or ChatGPT the script and every asset, each with a label, and have it write one structured prompt per scene. Abhinav Kalyan, Estefania Freites, @I_Muhammadali44 and @commonman_16 all do this. Abhinav Kalyan uses the same asset labels in Claude and in Higgsfield, so the prompt's references attach to the right files. When a shot misses, he describes what the camera actually shows and asks Claude to rewrite the prompt.

@lr_mvcreative recommends saying, for each reference image, what to take from it and what to ignore.

5. Set the look with lens, grain and focus terms

Describe the camera as a cinematographer would. Abhinav Kalyan adds an anamorphic lens and film grain to remove a plastic digital look. Ayush shore uses an f/1.4 shallow depth of field so the background blurs and stays stable. Alber Adilov renders at 4K with high bitrate to keep smoke and deep reds clean.

6. Run several takes per scene and keep the best

Generate each scene more than once and cut from the best moments. Abhinav Kalyan runs four or more batches per scene. Janvi Shah treats choosing frames as the main directing job. Alber Adilov's summary notes that the CGI shot in the guide he shares took two tries. Abhinav Kalyan describes his result plainly: "The final film is the best 30 seconds of 100 tries, cut together."

Budget for this. A commercial costs several times its length in generations.

7. Finish sound, grade and assembly

Sound is where the sources disagree most. Manar Elmasry adds an AI voiceover and music matched to the mood. Ayush shore uses one instrumental track throughout. @lr_mvcreative uses no music at all, only room tone and action sounds. Pick one approach per spot and hold it.

Assemble in an editor, fix any reversed shots, then check: same actor and product in every scene, no generation errors, no claims the brand cannot make, and the platform's AI label rules.

Prompt template for a commercial scene

This template is ours. It follows the Higgsfield Help Center's shot-list layout (blocks for style and mood, camera, light and colour, action, audio, and shots) and, inside each shot, the BytePlus Seedance order: camera move, subject action, position, then audio. Seedance's guide advises one camera move per shot and labelled shots in order rather than per-shot timings. See Higgsfield prompts for more, or build one in the AI ad prompt builder.

References: @Character = [name] sheet (face and full body). @Product = [product] sheet (front, three-quarter, back). @Location = [place] sheet.
Style and mood: [genre] commercial, [mood], anamorphic lens, light film grain.
Light and colour: [key light], [palette], [time of day].
Shot 1: slow dolly in. @Character stands at [position in @Location], looks up from [action]. Ambient sound of [place].
Shot 2: static close-up. @Character's hand lifts @Product toward the lens; the label faces the camera.
Shot 3: low-angle wide, slow pan right. @Character walks out of frame; @Product stays sharp in the foreground. <[sound effect]>
Shot 4: static hero shot of @Product on [surface], empty space above for the end card.
Keep @Character's face, hair and outfit the same in every shot. Keep @Product's shape and label exactly as in its sheet. No visible crew, camera gear or drones. Keep it subtitle-free. Do not generate a watermark.

Common problems and fixes

What creators report

These are the authors' own claims.

• Estefania Freites reports producing 12 fifteen-second 4K ads a week with the production board method.
• Abhinav Kalyan reports that his 30-second film was cut from about 100 tries.

Anglr pays AI creators and editors on performance to remake winning scripts as brand films and other formats; see Anglr for creators.