How to make UGC-style video ads with AI: workflow and prompts (2027)

You want a UGC-style ad, but an AI presenter should not create a product experience that never happened. Before choosing the actor or generating footage, the audience problem and substantiated claim need to be clear. Otherwise a polished performance can say the wrong thing convincingly.

This guide builds the ad from that brief. You’ll see how to write the spoken script, supply the product image and separate visual directions, then review the entire result. Test different hooks against useful customer outcomes while keeping synthetic actors distinct from real testimonials.

What the current Magic Hour workflow accepts

An AI actor is not a customer testimonial. Do not give a synthetic presenter a personal product experience, result or endorsement that did not happen. Disclose AI-generated media where the platform or law requires it.

The current Magic Hour AI UGC Ad Generator lets you choose a preset actor or upload an authorized actor image, optionally add a product image, enter spoken words in Script, and enter visual directions in Action prompt. Use the signed-in editor for these controls; the simplified public form is a different entry point. Available settings and the credit estimate can change with the selected options.

Magic Hour’s current product guide separates the required actor image, optional product image, spoken script and visual directions. To use an authorized recording, choose Upload script audio when the current Script controls offer it and check the selected audio section. This option is model-specific; do not assume every model accepts uploaded audio. Recheck the controls, duration and estimate after changing the selection.

Watch a shot prompt become a video workflow

This official Magic Hour walkthrough shows a text-to-video workflow, shot prompts and reusable references. Follow how the brief becomes a shot, then check the complete result. It is an instructional recording, not a test of every prompt or model in this article.

Source: Magic Hour AI: Easiest Guide to Mastering Text-to-Video with Perfect Consistency | Magic Hour AI

Prepare a brief before generating

• Audience and moment: identify who sees the ad and what just happened before they encounter it.

• One problem: describe the concrete job or friction the product addresses.

• One claim: use wording supported by product behavior, customer evidence or a documented test.

• Proof: decide what the viewer can actually see: interface recording, product close-up, demonstration or verified customer result.

• Action: specify the next step, such as view the product, start a free test or request a demo.

• Rights and disclosure: confirm permission for the actor, product, voice, music and claims before production.

Seven-step AI UGC ad workflow

1. Write three meaningfully different hooks

Change the reason to keep watching, not a single adjective. Useful hook families include a direct problem, a surprising demonstration and a comparison with the current workflow.

Prompt: “Write three 8–12 word hooks for [audience] who struggle with [problem]. Each hook must introduce a different angle. Use only these supported facts: [facts]. Do not invent a testimonial, result or statistic.”

2. Turn one hook into a short factual script

Use a simple sequence: hook → problem → visible proof → supported benefit → action. Read it aloud and remove claims that the footage cannot support. A synthetic actor can explain or demonstrate a product; it should not pretend to be an actual customer.

Prompt: “Write a 20-second presenter script using this hook: [hook]. Audience: [audience]. Show this proof: [proof]. Supported claim: [claim]. End with [action]. Use spoken language. Do not add personal experience, customer results, urgency or guarantees.”

3. Prepare the actor and product images

Use a clear actor image you have rights to use. If the product must remain recognizable, upload a clean product image and inspect labels, logos, shape and color in the result. The current Magic Hour guide documents several image formats, an 8,000-pixel side limit and plan-dependent upload thresholds.

4. Separate speech from action

Put only words the presenter should say in Script. Put camera, gesture, setting and movement instructions in Action prompt. Mixing dialogue with visual directions can cause missing speech or literal on-screen behavior.

Use the example below as an Action prompt in the UGC editor after choosing the actor, optional product and spoken Script. A matching script template is: “Meet [product]. Here is [visible, verified detail]. Check the current options on the product page.” Replace the placeholders with substantiated details. This is an untested input example, not a generated result or customer testimonial. For a new scene that does not use this actor-and-product ad form, open Text-to-Video instead.

5. Choose settings and generate

Select the model, duration, aspect ratio and resolution available in the editor. Recheck the displayed credit estimate after every model or duration change. Generate one diagnostic version before creating a batch.

6. Review the complete ad

• Claim accuracy: every spoken and visual claim matches the product and evidence.

• Identity and rights: the actor, voice, music, product and background are authorized.

• Product fidelity: labels, packaging and demonstrated behavior have not changed.

• Audio and captions: names, numbers and timing are correct.

• Disclosure: AI generation and material relationships are disclosed as required.

• Delivery: aspect ratio, safe areas, resolution and watermark fit the destination.

7. Launch a controlled creative test

Hold the audience, offer, landing page, budget and optimization event as stable as practical. Change one major creative variable per test—usually the hook, proof or actor—so the result is interpretable. Use platform experiments when available.

Track spend, qualified clicks, completed sign-ups, purchases, revenue and refund or cancellation quality. A higher watch rate with fewer purchases is not a winning revenue result. Record the creative, dates, audience, spend and conversion window before declaring a winner.

Claims and disclosures that protect trust

The FTC advertising guidance says endorsements must reflect the endorser’s honest experience and that advertisers need substantiation for claims. It also warns that atypical results require clear information about what consumers can generally expect; a vague “results may vary” disclaimer is insufficient.

TikTok Ads Manager guidance currently lists a disclaimer as mandatory for AI-generated, synthetic or significantly manipulated media. Policies vary by platform and region, so check the live rules for every destination before launch.

• Synthetic presenter: identify the media as AI-generated when required and never imply the actor is an actual customer.

• Performance claim: keep the evidence, population, method and date that support it.

• Before-and-after: show a representative process and disclose material conditions.

• Scarcity or price: verify the offer is live for the audience and dates shown.

• Customer quote: use the real words, permission and material-connection disclosure.

A practical variation matrix

• Hook: problem, demonstration or comparison.

• Proof: interface capture, product close-up or verified result.

• Presenter: one authorized actor per variant.

• Setting: change only when it expresses a different use context.

• Action: keep stable unless CTA wording is the test.

Start with three hooks against one script body and one offer. If one hook produces enough conversion volume to evaluate, test the proof or presenter next. Generating dozens of variants without enough spend per cell creates noise rather than learning.