How to make AI UGC ads

What is an AI UGC ad?

An AI UGC ad is a short vertical video that looks like a real customer filmed it on a phone, but the person, scene and motion are generated by AI. It usually shows one creator talking to camera, holding or using the product, with captions and a call to action.

Common formats are the selfie review, the get-ready routine, the reaction hook and the talking head. You can see public examples of AI ads in these formats on AI get-ready UGC ads and AI talking head ads. Those examples come from posts on X; they were not made on or for Anglr.

Tools creators use for it

The table counts how many of the reviewed tutorials used each tool for UGC ads. A tutorial that used two tools counts once for each.

Based on 40 public tutorials posted between 20 Jan 2026 and 7 Oct 2026, reviewed by Anglr.

How do you make an AI UGC ad step by step?

You make it in seven steps: script, creator, product reference, start frame, clips, voice, edit. The order below is our synthesis of the tutorials; each step notes where the sources agree and where they differ.

1. Start from a script and a proven angle

Write the hook, the problem, the product moment and the call to action before you open a video tool. At least five authors, including @EXM7777, @petersapps and Kalash Jain, research ads that are already running in the category and rewrite the structure for their own product.

Sources differ on where the script comes from. Some write it by hand; most hand the product and angle to Claude or ChatGPT and edit what comes back. If you work from a brief, the angle and the winning script are already chosen, which is how Anglr briefs work (see Anglr for creators). A plain UGC script structure also helps.

2. Design one creator and lock the face

Make one creator image, then a reference sheet of that face. This is the step the most sources stress, because face drift between clips is the most common failure.

Aswad Tahir uses a front, three-quarter and profile sheet. Eugenio Zabell fixes face, hair, wardrobe and light on one sheet and adds a clause that the face must match. @hamzaoudghiri2 saves the character as a Higgsfield Soul ID instead. All three aim at the same result: one face the video model sees every time.

Several authors also pull the look from a real reference rather than inventing it. @brainextends, George Stock and @FynCas start from a real UGC video or a Pinterest image and copy its framing, light and setting, not the person.

3. Use a clean photo of the real product

Attach a sharp, straight-on photo of the real product with a readable label. At least four authors do this so the packaging in the ad matches what the customer receives. Without it, the models tend to invent label text.

4. Generate a phone-style start frame

Make the first frame as a still: the creator, in the setting, holding the product, shot like a selfie or a propped-up phone. Then animate that still. Nearly every tutorial uses image-to-video rather than text-to-video for UGC, because the still fixes the face, the product and the light before any video credit is spent.

Estefania Freites puts the reason in one line: "Images are cheap, video is expensive." Iterate on the still until it is right.

For realism, Tyler Reed and @sixugc ask for a selfie angle, natural skin texture and a handheld feel instead of studio polish.

5. Generate short clips, one action at a time

Generate the ad as short clips or timed beats, not one long take. @eptwts splits the ad into 4 to 8 second clips, each with its own start frame. Eugenio Zabell controls each 4 to 5 second block and chains them. @maxxmalist writes one prompt per scene.

Here the sources disagree with the model vendor. Many creators write timestamps into the prompt. BytePlus's Seedance 2.0 prompt guide says per-shot timings are unstable and recommends labelling shots in order instead. Our template below follows the vendor; if timestamps work for you, keep them.

Estefania Freites goes the other way and renders 30 seconds in one Seedance run, then regenerates only the weak scenes from their own start frames. That is one author's method; most sources work clip by clip.

6. Keep the voice the same

Decide the voice once. Will Connell makes the voiceover separately in ElevenLabs. Eugenio Zabell attaches a voice reference to the generation. Whichever route you take, use the same voice in every clip.

7. Edit, caption and check

Cut the clips, add captions, speed up slow moments and add music in an editor. @codeMdSanto, Peter Njoagwuani and Mike Futia all finish in CapCut; none of them treats the raw generation as the final ad.

Before it goes out, check the product looks the same in every shot, there are no visible generation errors and every claim is one the brand can legally make. On TikTok, AI-generated ads need the AIGC label or a clear disclaimer of your own (TikTok ad policy).

Make variations from the same script

Once one version works, change one thing at a time: the face, the setting, the tone or the hook. @davidfigeira, George Stock and Abhinav Kalyan all build several versions from one script so the test compares like with like. One script, every format covers the same idea across formats.

Prompt template for an AI UGC clip

This template is ours. It follows the order in BytePlus's Seedance guide (subject, action, setting, light, camera, style, dialogue, audio), with dialogue in curly braces and constraints at the end. Duration and aspect ratio go in the settings, not the prompt. Build your own version in the AI ad prompt builder or see more on Seedance prompts.

Start frame (image model):

A vertical phone selfie photo for a [product category] ad. A [age] [gender] with [two or three stable features: hair, top, accessory] holds [product] at chest height, label facing the camera and readable. Setting: [room], [time of day] window light. Framing: arm's-length selfie, slight high angle, natural skin texture, everyday clothes, no studio lighting.

Motion (Seedance, image-to-video):

Image 1 is the start frame. Image 2 is the creator reference sheet. Image 3 is the product.
Shot 1: handheld selfie, small natural drift. The woman from Image 2 looks into the lens and raises [product] slightly. She says {[hook line]}.
Shot 2: same framing. She turns the label toward the camera and taps it once. She says {[benefit line]}.
Shot 3: same framing. She smiles and nods. She says {[call to action]}.
Audio: room tone, no music.
Keep the face, hair and outfit the same as Image 2. Keep the product shape and label the same as Image 3. Keep it subtitle-free. Do not generate a logo or a watermark.

Common problems and fixes

What creators report

These are the authors' own claims, not results we checked.

• Mike Futia reports shipping an ad in about 20 minutes with his Claude Code setup.
• @maxxmalist reports making a pharmacist-style UGC ad in 5 minutes in Kling 3.0.
• @jacobrodri_ reports that an app using the reaction-then-demo format reached 2.2 million TikTok views.
• Estefania Freites reports making 12 multi-scene ads in a week with a single production board per ad.

If you already make AI UGC ads that perform, the creators page explains how Anglr pays for them, and the UGC creator guide covers the basics if you are starting out.