3 minutes. 3 tools. that's all it took for me to make this video > sora 2 > V2 prompting system GPT > optional - topaz upscale you can create AI UGC videos for literally anything. software, ecom, consumer apps, info products, whatever industry you're in... this is probably https://t.co/9srmOr2rC7

Miko

@Mho_23 · creator

TWITTER · 2026-03-09Source

119likes

15comments

Remix This

Prompt

GLOBAL LOCK:
Subject: Young Caucasian woman, mid-20s, dark brown hair tied back in a messy bun, fair skin with natural texture and warm undertones, slim build.
Wardrobe: Simple grey ribbed tank top, thin gold chain necklace.
Environment: Domestic bathroom, white tiled walls, large mirror with a bright overhead light causing a soft glare. Chrome faucet and toiletries (toothpaste, brown pump bottle) visible on the counter.
Lighting: Warm, slightly yellow indoor lighting, soft shadows, realistic glare from the mirror.
Color Grade: Natural, warm, slightly desaturated, UGC/smartphone video aesthetic.
Camera: Handheld, slight natural jitter, 35mm lens feel, shallow depth of field.
Speech: Single female speaker, casual UGC style, warm and sincere tone, medium pace, clear enunciation with natural pauses and filler words.

[00:00–00:03]
Subject: Woman looks slightly off-camera then directly at the lens.
Action: She begins speaking: "Okay, so like...". Expression is thoughtful and slightly hesitant.
Camera: Medium Close-Up (MCU), handheld jitter.
Speech: "Okay, so like," (hesitant tone).
Sync: High lip-sync strictness.

[00:03–00:07]
Subject: Woman looking into the mirror reflection then back to the camera.
Action: She raises her right hand and gently touches her cheek while saying: "I don't even know if I should say this...".
Camera: MCU, slight movement following her hand.
Lighting: Glare on the mirror is prominent.
Speech: "I don't even know if I should say this," (confessional tone).
Sync: High lip-sync strictness.

[00:07–00:11]
Subject: Woman gestures with both hands, palms open.
Action: She says: "but my skin was like bad for months. Like I tried everything...". Expression is earnest and frustrated.
Camera: MCU, handheld movement.
Speech: "but my skin was like bad for months. Like I tried everything," (emotive tone).
Sync: High lip-sync strictness.

[00:11–00:13]
Subject: Woman leans in closer to the camera.
Action: She says: "and the only thing, and I'm being so serious right now, was beef tallow."
Camera: Close-Up (CU), intimate framing, focus on eyes and mouth.
Speech: "and the only thing, and I'm being so serious right now, was beef tallow." (sincere, emphatic tone).
Sync: High lip-sync strictness.

[00:13–00:15]
Subject: Woman smiles slightly and gives a firm nod.
Action: She says: "I know, I know how that sounds. I don't know, but this stuff really works." She waves her hand dismissively.
Camera: CU, direct eye contact with the lens.
Speech: "I know, I know how that sounds. I don't know, but this stuff really works." (convinced, reassuring tone).
Sync: High lip-sync strictness.

NEGATIVE PROMPT:
Visual: AI artifacts, flickering, unnatural skin texture, distorted fingers, background warping, cinematic lighting, high saturation, professional studio look, static camera, perfect symmetry.
Speech: Robotic voice, monotone delivery, misaligned lip-sync, background noise, harsh sibilance, unnatural breathing, studio-dry audio.

SPEECH PACK:
Transcript: "Okay, so like, I don't even know if I should say this, but my skin was like bad for months. Like I tried everything and the only thing, and I'm being so serious right now, was beef tallow. I know, I know how that sounds. I don't know, but this stuff really works."
TAKE_A: Conversational and sincere, with natural pauses after "like" and "this."
TAKE_B: Slightly hesitant at first, then gaining confidence when mentioning the product.
TAKE_C: Direct and persuasive, emphasizing "so serious" and "really works."
Prosody: [pause] after "like", [emphasis] on "bad", [emphasis] on "serious", [slight laugh] on "I know how that sounds".

How Mho_23 Made This AI UGC Skincare Testimonial Video and How to Recreate It

This case study analyzes a groundbreaking example of "Stealth AI" content—a 15-second skincare testimonial that looks indistinguishable from a standard iPhone UGC (User Generated Content) video. The video features a young woman in a domestic bathroom setting, sharing a sincere recommendation for "beef tallow" skincare. By utilizing Sora 2 and a V2 prompting system, the creator achieved a level of realism that bypasses the "AI uncanny valley." The core keywords here are cinematic UGC, bathroom lighting, handheld aesthetic, and high-trust testimonial. For indie creators, this represents a shift from "AI as a gimmick" to "AI as a production tool" for high-converting social media content.

Visual Breakdown: The "Stealth AI" Aesthetic

The video's power lies in its imperfections. We see a young woman with dark hair tied back, wearing a simple grey tank top—the universal uniform of a relatable creator. The setting is a bathroom, complete with a mirror reflection, a chrome faucet, and everyday items like toothpaste. The lighting is intentionally "bad" in a realistic way: a warm, slightly yellow overhead light that creates a soft glare on the mirror and natural shadows on the subject's face. The camera movement mimics a handheld phone, with slight jitters and a focus that feels organic rather than digitally perfect. The skin texture is detailed, showing natural pores and subtle imperfections, which is crucial for a skincare-related video.

Shot-by-Shot Breakdown

Time Range	Visual Content	Shot Language	Lighting & Tone	Viewer Intent
00:00–00:03	Woman looks at camera, slightly hesitant.	Medium Close-Up (MCU), handheld.	Warm bathroom light, soft glare.	The Hook: "Is she about to tell a secret?"
00:03–00:07	She touches her cheek while talking.	MCU, focus on hand-to-face contact.	Natural skin tones, warm shadows.	Establishing Trust: Physical proof of skin quality.
00:07–00:11	Leans in closer to the camera.	Close-Up (CU), intimate framing.	Increased detail on facial features.	Reinforcing Sincerity: "I'm being serious."
00:11–00:15	Smiles and nods, looking directly at lens.	CU, direct eye contact.	Consistent warm domestic vibe.	The CTA/Closing: Final social proof.

The Viral Mechanism: Why This Works

The "Confessional" hook is the primary driver here. By starting with "I don't even know if I should say this," the creator triggers a psychological curiosity gap. Humans are biologically wired to pay attention to secrets or controversial advice. The choice of "beef tallow" as the product is brilliant—it's a niche, trending topic that sits at the intersection of "ancestral health" and "modern skincare," often sparking debate in the comments which further boosts the algorithm.

Platform Signals & Watch Time

From a platform perspective (TikTok/Reels/X), this video excels because it doesn't look like an ad. The 0–3 second hook is purely emotional and visual, preventing the "scroll reflex." The pacing is conversational, with natural pauses and "um"s that signal authenticity. Because the video is short (15 seconds) and high-value, it likely achieves a high completion rate and "loop effect," signaling to the algorithm that the content is worth pushing to a wider audience.

5 Testable Viral Hypotheses

The "Secret" Hook: Starting with a disclaimer of hesitation increases retention by 20% compared to direct product mentions.
The "UGC Mirror" Effect: Filming in a bathroom mirror increases trust scores because it mimics how real people share advice.
The "Niche Controversy" Play: Using a polarizing ingredient (like beef tallow) generates 3x more comments than generic ingredients like "aloe vera."
The "Handheld Jitter" Hypothesis: Adding 5% camera shake to AI video reduces the "AI perception" by users, leading to higher shares.
The "Skin Texture" Detail: High-resolution skin pores in a skincare video are the primary "proof" signal that stops the scroll.

How to Recreate: From 0 to 1

Topic Selection: Choose a "high-trust" niche like skincare, finance, or productivity where personal testimonials carry weight.
Scripting for Realism: Write a script that includes "filler words" (like, so, um) and a "hesitation hook."
Character Consistency: Use a consistent seed or reference image for your AI model to ensure the face doesn't shift between shots.
Environment Prompting: Specifically prompt for "domestic bathroom," "overhead warm light," and "mirror reflection" to ground the character in reality.
Motion Generation: Use tools like Sora 2 or Kling to generate segments. Focus on hand-to-face movements and natural eye blinks.
Audio Syncing: Use an AI voice cloner (like ElevenLabs) with a "casual/conversational" preset, then use a lip-sync tool (like SyncLabs or Hedra) to match the video.
Post-Processing: Use Topaz Video AI to upscale and add a slight "film grain" or "digital noise" to mimic a smartphone camera sensor.
Publishing: Post with a caption that asks a question (e.g., "Has anyone else tried this?") to spark engagement.

Maintaining Character Consistency

To keep the same person across multiple videos, create a "Character Sheet" prompt. Describe the subject's ethnicity, specific facial features (e.g., "slight downward tilt to the eyes"), and hair texture. Use the same "Global Lock" parameters in every generation to ensure the wardrobe and lighting remain identical.

Growth Playbook: Hooks & Captions

Opening Hook Lines

"I was actually terrified to post this, but..."
"Stop using [Common Product] and do this instead."
"3 minutes. 3 tools. That's all it took."

Caption Templates

The "Secret Sauce" Template:
I don’t know who needs to hear this, but [Product] changed everything for me. 🤫 I tried the expensive stuff for months with zero results. This is my 30-day update. Have you guys heard of this? 👇 #skincaretips #ugccreator

Hashtag Strategy

Broad: #skincare #beauty #wellness (To reach a wide audience)
Mid-tier: #beeftallow #naturalskincare #ugc (To hit the specific niche)
Niche: #aisora #aiugc #ancestralhealth (To capture tech-savvy or hyper-niche users)

Frequently Asked Questions

What tools make it look the most similar?

Sora 2 for video generation and ElevenLabs for the conversational voice clone.

What are the 3 most important words in the prompt?

"Handheld," "Domestic," and "Natural Lighting."

Why does the generated face look inconsistent?

You likely aren't using a "Global Lock" on facial features or a consistent seed value.

How can I avoid making it look like AI?

Add digital noise/grain in post-production and ensure the character touches their face or hair.

Is it easier to go viral on Instagram or TikTok?

TikTok favors this "raw" UGC style more than the polished aesthetic of Instagram.