Learning Objectives
By the end of this module, you will be able to:
- Choose the right AI UGC platform based on your use case and budget
- Write ad scripts that sound human, not AI-generated
- Direct AI avatars for natural delivery and authentic body language
- Apply the "imperfection stack" to make AI UGC indistinguishable from organic content
- Produce a complete UGC-style testimonial ad from script to export
Two Approaches to AI UGC
There are two fundamentally different methods for producing UGC-style AI video ads. Each has distinct strengths.
Approach A: Avatar Platforms (Fastest)
Platforms like HeyGen, Arcads, and Creatify generate video of AI spokespersons delivering scripted lines. You write (or generate) a script, select an avatar, choose a voice, and the platform produces a complete talking-head video.
HeyGen — the market leader in AI avatar video. Features 200+ stock avatars with natural movement and expression. Supports voice cloning (upload 2 minutes of audio to create a synthetic copy). Key feature: "Interactive Avatar" mode produces more natural gesture and expression than static avatars.
Arcads — purpose-built for ad creative. Generates UGC-style videos optimized for performance advertising. Stronger "authentic" feel than HeyGen because avatars are specifically trained for testimonial-style delivery. Includes script generation, A/B variant creation, and direct platform integration.
Creatify AI — the "product URL to ad" pipeline. Paste a product URL, Creatify scrapes the page, generates scripts, creates avatar videos, and outputs platform-ready ads. Lowest effort for e-commerce brands with existing product pages.
When to use avatar platforms: You need volume (10+ variants fast), you don't have time for custom AI video production, and the content is testimonial/talking-head format.
Approach B: Full AI Production (Highest Quality)
Generate photorealistic "UGC-style" imagery with Nano Banana Pro or Midjourney, then animate with Veo 3.1 or Kling, adding voice via ElevenLabs. This produces output that's visually indistinguishable from real UGC but requires more time and skill.
When to use full production: Quality matters more than speed, you need specific brand aesthetics or demographics not available in avatar libraries, or your product requires physical demonstration that avatar platforms can't simulate.
Writing Scripts That Sound Human
The #1 failure point in AI UGC is the script. Generic AI-generated scripts produce generic AI-sounding ads. The fix is structured specificity.
The "Real Person" Script Formula
Every effective UGC testimonial follows this structure:
HOOK (0-3 seconds)
Something that stops the scroll — a bold claim, a relatable frustration,
or an unexpected visual.
PROBLEM (3-8 seconds)
The specific pain point the viewer identifies with.
Must feel personal, not like marketing copy.
DISCOVERY (8-12 seconds)
How the speaker found the product. Casual, almost accidental.
Never "I was researching solutions for..."
PROOF (12-20 seconds)
Specific, tangible result. Numbers, timeframes, sensory details.
"My skin cleared up in two weeks" not "It really works."
CTA (20-25 seconds)
Direct, low-pressure. "Link's in the bio" not "Click now for 50% off."
Prompting for Non-Generic Scripts
Feed this framework to ChatGPT or Claude with product-specific context:
Write a 25-second UGC testimonial script for [product name],
a [product category] that [core benefit].
The speaker is a [age/gender/lifestyle] who [relevant context].
Requirements:
- Opening hook must be unexpected or emotionally specific
- Include at least ONE moment of hesitation or self-correction
(e.g., "I wasn't sure at first, honestly" or "It sounds weird but—")
- The discovery moment should feel casual and accidental
- Include ONE specific, measurable result with a timeframe
- End with a soft CTA (not hard sell)
- Total word count: 65-80 words (25 seconds at natural speech pace)
- Conversational, slightly rambling, not polished
DO NOT use these phrases: "game changer," "I have to share,"
"you guys," "obsessed," "literally," "amazing product,"
"I can't believe," "link in bio"
Example Output (Skincare Brand)
Bad script (generic AI):
"Hey guys! I just discovered this amazing new serum and I'm literally
obsessed. It's been a total game changer for my skin. I can't believe
the results. You have to try this. Link in bio!"
Good script (directed AI):
"Okay so my dermatologist actually told me to stop using actives for
a month. And I was panicking because — well, you know. So I switched
to just this barrier repair serum from [brand]. Nothing else. And I
kid you not, by week three my texture was... genuinely better than
when I was using five products. The irony. Anyway, I'll drop the
link if you want to check it out."
The second script works because it has a specific scenario (dermatologist advice), emotional authenticity (panicking), self-correction ("well, you know"), a specific timeframe (week three), an unexpected insight (less is more), and casual language that sounds spoken, not written.
Script Variant Generation
Once you have one strong script, generate 10-20 variants with different hooks:
Take this testimonial script and create 10 alternative opening hooks.
Each hook should be 5-8 words and use a different emotional angle:
1. Frustration ("I was so tired of...")
2. Surprise ("I didn't expect this to actually...")
3. Social proof ("My friend wouldn't shut up about...")
4. Confession ("Okay, embarrassing admission...")
5. Challenge ("I was skeptical when...")
6. Contrast ("I used to spend $200 a month on...")
7. Question ("Has anyone else noticed that...")
8. Urgency ("I found this right before it sold out...")
9. Result-first ("Three weeks. That's all it took.")
10. Anti-marketing ("This isn't sponsored, I just—")
Keep the same body and CTA. Only change the first 2 sentences.
This gives you 10 hook variants from a single base script — the foundation of multi-variant testing (covered in Module 4).
The Imperfection Stack: Making AI UGC Look Real
Real UGC has specific visual characteristics that signal authenticity. AI-generated content tends to be "too perfect" — too well-lit, too centered, too clean. The imperfection stack deliberately adds the markers of authentic amateur content.
Layer 1: Environment Imperfections
When generating your setting (via Nano Banana Pro or as an avatar platform background):
Authenticity markers to include:
- Rumpled bedding or unmade bed visible in background
- Cluttered nightstand (water bottle, phone charger, lip balm)
- String lights or fairy lights (signals "real bedroom")
- Personal photos on wall in random arrangement
- Plush toy or stuffed animal visible
- Slightly wrinkled clothes on a chair
- A visible laundry basket (partially full)
- Real books or magazines stacked imperfectly
Prompt for Nano Banana Pro UGC background:
"A lived-in young woman's bedroom, natural daylight from a window
on the left. Unmade bed with beige linen duvet and scattered pillows.
Nightstand with a glass of water, phone charger cable, and a small
plant. String lights along the headboard. A few personal photos
taped to the wall. The room is tidy but not perfect — it looks real,
like someone actually lives here. iPhone photo quality."
Layer 2: Camera/Framing Imperfections
Real UGC is shot on phones, often propped up on surfaces. The framing reflects this.
Camera characteristics of authentic UGC:
- Slightly low angle (phone propped on desk or shelf)
- Subject not perfectly centered (offset left or right)
- Slight tilt (1-3 degrees off horizontal)
- Auto-exposure adjustments visible (slight brightness shifts)
- Front-facing camera quality (softer than rear camera)
- Occasional glance down at phone (checking they're recording)
For Veo 3.1 prompts, include: "Shot from a low angle as if the phone is propped on a desk. The speaker is slightly off-center. Front-facing smartphone camera quality — slightly soft focus, natural auto-exposure."
Layer 3: Subject Imperfections
The person shouldn't look like a model in a studio.
Subject authenticity markers:
- Natural skin texture (visible pores, slight unevenness)
- Minimal or no visible makeup (or "no makeup" makeup)
- Slightly wrinkled or casual clothing (oversized t-shirt, hoodie)
- Hair that's been styled casually (messy bun, air-dried)
- Natural gestures (touching hair, adjusting position)
- Brief eye breaks from camera (glancing at product, looking away thinking)
- Genuine micro-expressions (brief frown, eyebrow raise)
Layer 4: Audio Imperfections
Perfect audio screams "produced." Real UGC has environmental audio.
Audio authenticity markers:
- Slight room reverb (not a treated studio)
- Background noise (distant traffic, AC hum, dog barking)
- Phone microphone quality (slightly compressed, not crystal clear)
- Natural speech patterns (um, uh, brief pauses, self-corrections)
- Slightly uneven volume (moving closer/farther from phone mic)
For ElevenLabs, reduce the "stability" slider to 0.4-0.5 (introduces more natural variation) and reduce "clarity" slightly for phone-mic quality.
Platform-Specific AI UGC Workflows
HeyGen Workflow (Avatar)
Step 1: Select Avatar
→ Browse library → Choose based on target demographic
→ "Interactive Avatar" mode for most natural movement
Step 2: Input Script
→ Paste your directed script
→ Select voice (or upload voice clone audio)
→ Set speech pace: slightly faster than default (people speak faster in UGC)
Step 3: Set Scene
→ Upload custom background (your Nano Banana Pro bedroom/bathroom)
→ Or use HeyGen's stock backgrounds
→ Ensure background matches UGC aesthetic (casual, lived-in)
Step 4: Generate & Review
→ Check lip sync accuracy
→ Check gesture naturalness
→ Check eye contact consistency
Step 5: Post-Process
→ Add platform-native elements: captions, product B-roll cuts, end card
→ Export at 9:16 for TikTok/Reels, 4:5 for feed
Creatify AI Workflow (URL-to-Ad)
Step 1: Paste Product URL
→ Creatify scrapes product images, descriptions, reviews
Step 2: Select Ad Format
→ UGC Testimonial / Product Demo / Problem-Solution / Comparison
Step 3: Review & Edit Script
→ Creatify generates a script — ALWAYS edit it
→ Apply the "Real Person" formula from above
→ Remove any generic marketing language
Step 4: Select Avatar + Voice
→ Match to target demographic
→ Custom voice if available
Step 5: Generate Variants
→ Create 5-10 variants with different hooks
→ Export all for A/B testing
Full Production Workflow (Nano Banana Pro + Veo 3.1)
For maximum quality and brand control:
Step 1: Generate Character
→ Nano Banana Pro: Create a realistic "everyday person" character
→ Match target customer demographic precisely
→ Generate in casual attire, natural environment
Step 2: Generate UGC-Style Keyframe
→ Combine character + imperfect bedroom/bathroom environment
→ Phone-propped low angle composition
→ Natural lighting, casual pose
Step 3: Generate Video in Veo 3.1
→ Upload keyframe as start image
→ Include dialogue in prompt with casual delivery direction
→ "She speaks directly to the phone camera in a conversational,
slightly rambling tone. She occasionally looks down at the
product in her hand. Natural room acoustics."
Step 4: Generate Voice Separately (ElevenLabs)
→ Reduce stability for natural variation
→ Slightly breathy, conversational delivery
Step 5: Composite in Post
→ Layer ElevenLabs audio over Veo video (or use Veo native audio)
→ Add captions (large, centered, TikTok-native style)
→ Add product B-roll cut (2-3 seconds of product close-up)
→ Export at 9:16
Practical Exercise
Exercise: Produce a 30-Second UGC Testimonial Ad
Choose a product (your own brand or a fictional one). Then:
- Write the script using the "Real Person" formula (hook, problem, discovery, proof, CTA). Use ChatGPT/Claude with the prompting template from this module.
- Generate the environment in Nano Banana Pro — a lived-in bedroom or bathroom with imperfection markers.
- Produce the video using either:
- Option A: An avatar platform (HeyGen free tier, Creatify free trial)
- Option B: Full production (Nano Banana Pro character + Veo 3.1 animation)
- Add platform elements: large captions, a 2-second product B-roll cut, end card with CTA
- The authenticity test: Show the finished ad to someone who doesn't know it's AI-generated. Do they flag it as AI within the first viewing?
Key Takeaways
- Two approaches to AI UGC: Avatar platforms (HeyGen, Creatify, Arcads) for speed and volume, or full AI production (Nano Banana Pro + Veo) for maximum quality and control.
- Script quality determines ad quality. Use the "Real Person" formula and explicitly prompt for hesitation, specificity, and anti-marketing language.
- The imperfection stack (environment, camera, subject, audio imperfections) is what makes AI UGC pass the authenticity test.
- Generate 10+ hook variants from a single base script — creative testing velocity is the AI advantage.
- The 2-second rule: If a viewer identifies the content as AI within 2 seconds, the ad fails regardless of the message.
References & Resources
- HeyGen: heygen.com
- Creatify AI: creatify.ai
- Arcads: arcads.ai
- ElevenLabs: elevenlabs.io
- InVideo: Top UGC Tools for Viral Ad Videos
- Pinterest board — UGC Ad Examples: https://pinterest.com/search/pins/?q=ugc%20ad%20examples%20tiktok
- Pinterest board — Authentic Bedroom Aesthetic: https://pinterest.com/search/pins/?q=aesthetic%20bedroom%20real%20lived%20in