0:00 / 0:00

How to create freakishly realistic AI videos 👏🔥 Comment “AI” for the resources! Arcads is an AI UGC platform that generates ultra-realistic, creator-style ads using consistent AI avatars and products. This is a huge breakthrough for brands and marketers who want authentic-looking UGC videos at scale without hiring creators or filming content. #Arcads #AIUGC #AIMarketing #AIVideoAds #CreativeAI

Rourke Sefton-Minns

@rourke · creator

INSTAGRAM · 2026-01-17Source

14.7Klikes

13.1Kcomments

Remix This

Recreate with Veo3.1

Make your own AI viral video

Prompt

GLOBAL LOCK: The video is a vertical 9:16 split-screen composite. 
BOTTOM HALF (Locked): A Caucasian male creator in his 30s with a beard, wearing a white baseball cap and a black sweater with a large white flower/heart design on the chest. He is positioned in the center of the bottom frame against a plain, flat-lit neutral wall. He acts as a commentator, looking up at the top frame, using expressive hand gestures (pointing, clapping) and speaking directly to the camera. 
TOP HALF (Locked): A series of AI-generated UGC-style clips and UI screen recordings. The primary AI character is a young Caucasian woman in her 20s, with dark brown hair pulled back in a messy bun, light freckles. The environment is a cozy bedroom with natural, soft window lighting coming from the right, featuring a bed, a corkboard with photos, and a bookshelf with plants. The camera style for the AI woman is a handheld smartphone front-facing camera (MCU framing). 
SPEECH STYLE: The bottom creator speaks with an energetic, instructional, authoritative tone (medium-fast pace, crisp articulation, close-mic dry room sound). The top AI woman (when speaking) has a fast, enthusiastic, gossipy UGC tone (high energy, uses filler words, slight room reverb).

[00:00–00:22]
TOP HALF: The AI woman is wearing a long-sleeved green top. She is holding a clear plastic cup of iced coffee with a straw in her right hand. She speaks enthusiastically to the camera, gesturing slightly with the coffee. At 00:18, the shot cuts, and she is now holding a red box of Colgate toothpaste, pointing to it with her left hand while continuing to speak. Text overlays appear: "AI is getting too realistic 🤯" and a large block of text titled "Veo 3.1 Prompt".
BOTTOM HALF: The male creator looks up in amazement. He points both index fingers upwards towards the top frame. He maintains an expression of shock and excitement.
SPEECH/AUDIO: 
Top Speaker (AI Woman): "Oh my god guys, do you ever just stop and realize how far AI has actually come? Like, it's actually insane how normal this feels now. I remember when it used to be just little short clips, right? But now it's full videos. The face looks real, the voice stays kinda consistent, everything just works. And the craziest part? I can literally hold products, talk about them, move them around. Like I'm actually filming this myself." (Lip sync strictness: High).
Bottom Speaker: None.

[00:23–00:26]
TOP HALF: The frame splits vertically into two smaller side-by-side videos. Left side labeled "Original": A man with long hair applying makeup with a brush. Right side labeled "Kling 01": The AI woman (in the green top) mimicking the exact same makeup application motion with a brush.
BOTTOM HALF: The male creator claps his hands together once, then points directly at the camera, then points up again.
SPEECH/AUDIO:
Bottom Speaker (Male Creator): "Here's exactly how you can create the most realistic AI UGC creatives." (Lip sync strictness: High).

[00:27–00:32]
TOP HALF: A static graphic of a PDF document titled "GenHQ: Long-form UGC style videos" appears, showing workflow steps and images of the AI woman.
BOTTOM HALF: The male creator continues speaking directly to the camera, using hand gestures to emphasize his points.
SPEECH/AUDIO:
Bottom Speaker (Male Creator): "I'll even give you this PDF document giving you a detailed list of instructions of how to do it at the end of the video." (Lip sync strictness: High).

[00:33–00:38]
TOP HALF: A screen recording of a web UI (Arcads). The cursor clicks "Image" then "See more". A text prompt is typed into a box. The screen shows two generated image options (A and B) of the AI woman, now wearing a long-sleeved red top with a black bra strap visible, holding the iced coffee. Option B is selected.
BOTTOM HALF: The male creator points up with his right index finger, explaining the UI steps.
SPEECH/AUDIO:
Bottom Speaker (Male Creator): "To get started, go to Arcads and go to the image section and write in a prompt detailing what you want your character to look like." (Lip sync strictness: High).

[00:39–00:42]
TOP HALF: A screen recording of an image editing UI. A text overlay says "change the coffee to a labubu doll". The image of the woman in the red top updates; the coffee cup in her hand is replaced by a small, tan plush doll.
BOTTOM HALF: The male creator continues explaining, hands clasped together.
SPEECH/AUDIO:
Bottom Speaker (Male Creator): "Now you can select the image you like and change it using Google NanoBanana Pro." (Lip sync strictness: High).

[00:43–00:50]
TOP HALF: A screen recording of a video generation UI. The image of the woman in the red top holding the coffee is set as "Start Frame". The image of her holding the plush doll is set as "End Frame". 
BOTTOM HALF: The male creator gestures with both hands, explaining the start/end frame concept.
SPEECH/AUDIO:
Bottom Speaker (Male Creator): "Once you've got these two images, you can use it as a start and an end frame with Google Veo 3. Replace your start frame with the end frame and repeat the process." (Lip sync strictness: High).

[00:51–00:56]
TOP HALF: The generated video plays: The AI woman in the red top is holding the plush doll, smiling and talking. She then holds up a printed copy of the PDF guide. A large, bold cyan text overlay "AI" appears in the center of the top frame.
BOTTOM HALF: The male creator points emphatically at the cyan "AI" text overlay.
SPEECH/AUDIO:
Top Speaker (AI Woman): [Inaudible/background speech].
Bottom Speaker (Male Creator): "Comment AI to get the entire workflow in the PDF." (Lip sync strictness: High).

NEGATIVE PROMPT:
Visual: temporal jitter, flickering lighting, morphed hands, extra fingers, melting objects, inconsistent facial features on the AI woman, text corruption in UI screens, blurry text overlays, unnatural robotic body movements, mismatched split-screen alignment, creator's background changing.
Audio: robotic voice cadence, harsh sibilance, audio clipping, lip-sync mismatch, unnatural pauses, lack of room tone in UGC segments.

SPEECH PACK:
[00:00-00:22]
Speaker: AI Woman
Transcript: "Oh my god guys, do you ever just stop and realize how far AI has actually come? Like, it's actually insane how normal this feels now. I remember when it used to be just little short clips, right? But now it's full videos. The face looks real, the voice stays kinda consistent, everything just works. And the craziest part? I can literally hold products, talk about them, move them around. Like I'm actually filming this myself."
Prosody: Fast-paced, enthusiastic, slight vocal fry, emphasis on "insane", "full videos", and "craziest part".

[00:23-00:56]
Speaker: Male Creator
Transcript: "Here's exactly how you can create the most realistic AI UGC creatives. I'll even give you this PDF document giving you a detailed list of instructions of how to do it at the end of the video. To get started, go to Arcads and go to the image section and write in a prompt detailing what you want your character to look like. Now you can select the image you like and change it using Google NanoBanana Pro. Once you've got these two images, you can use it as a start and an end frame with Google Veo 3. Replace your start frame with the end frame and repeat the process. Comment AI to get the entire workflow in the PDF."
Prosody: Instructional, clear articulation, punchy delivery, emphasis on tool names ("Arcads", "NanoBanana Pro", "Veo 3") and the final CTA ("Comment AI").

Why rourke's Realistic UGC Avatar Went Viral - and the Formula Behind It

This video is a masterclass in demonstrating "freakishly realistic AI" through a highly engaging, split-screen reaction format. The core visual hook relies on an AI-generated UGC (User Generated Content) avatar—a young woman in a cozy, warm-lit bedroom—seamlessly holding and interacting with physical products like an iced coffee, a tube of Colgate toothpaste, and a plush toy. By placing the creator (@rourke) in the bottom half of the screen reacting to and breaking down the workflow of the AI video playing above him, the video bridges the gap between high-tech AI capabilities and accessible creator tools. The aesthetic leans heavily into the "iPhone front-facing camera" vibe, utilizing natural window lighting, subtle depth of field, and relatable wardrobe choices (casual green and red tops) to sell the illusion of reality. The ultimate goal is lead generation, utilizing a massive curiosity gap and a direct "Comment AI" call-to-action to distribute a PDF guide, resulting in over 13,000 comments.

2. What You're Seeing

The video utilizes a persistent vertical split-screen layout. The bottom half features the real human creator, wearing a distinctive black sweater with a white heart motif and a white baseball cap, set against a plain, neutral background. His role is to act as the guide and the audience surrogate, using expressive hand gestures (pointing up, clapping) and facial expressions (amazement, direct eye contact) to direct attention to the top half. The top half is the dynamic showcase area. It begins with a hyper-realistic AI avatar of a young woman in a bedroom setting. The lighting simulates soft, natural daylight coming from a window off-camera, casting gentle shadows that enhance the 3D realism of her face and the objects she holds. The camera language mimics a handheld smartphone, with slight, naturalistic movements. The video then transitions into screen recordings of software UIs (Arcads, an image editor, a video generator), visually proving the "how-to" aspect before ending with a strong visual text overlay prompting the viewer to comment.

Shot-by-Shot Breakdown

Time Range	Visual Content	Shot Language	Lighting & Color Tone	Viewer Intent
00:00 - 00:22	Top: AI woman in green top holding iced coffee, then Colgate. Bottom: Creator reacting and pointing up. Text overlays showing prompts.	Top: MCU (Medium Close Up), handheld smartphone feel. Bottom: MCU, static tripod.	Top: Warm, natural window light, cozy bedroom tones. Bottom: Flat, even studio lighting.	Hook the viewer with hyper-realistic AI interacting with physical objects; establish the creator as the authority.
00:23 - 00:26	Top splits again: Left shows original man applying makeup, Right shows AI woman applying makeup. Bottom: Creator explaining.	Top: MCU for both, static. Bottom: MCU, static.	Top: Natural indoor lighting. Bottom: Flat studio lighting.	Demonstrate the source-to-output capability of the AI model (Kling 01), proving it's not just a deepfake but a full generation.
00:27 - 00:32	Top: Graphic of a PDF guide "Long-form UGC style videos". Bottom: Creator talking directly to camera.	Top: Static graphic. Bottom: MCU, static.	Top: Bright, clean graphic design. Bottom: Flat studio lighting.	Introduce the lead magnet/value proposition early to prime the audience for the CTA.
00:33 - 00:42	Top: Screen recording of Arcads UI, generating an image, then using an inpainting tool to change a coffee cup to a plush toy.	Top: Screen capture, zooming in on UI elements. Bottom: MCU, static.	Top: Bright white UI backgrounds. Bottom: Flat studio lighting.	Demystify the process; show that the workflow is accessible and tool-based, building trust.
00:43 - 00:50	Top: UI showing "Start Frame" and "End Frame" setup for video generation. Woman in red top.	Top: Screen capture. Bottom: MCU, static.	Top: Bright UI. Bottom: Flat studio lighting.	Explain the specific technical trick (keyframing) used to achieve the consistent product interaction.
00:51 - 00:56	Top: AI woman in red top holding the plush toy, then holding the printed PDF guide. Large cyan "AI" text appears. Bottom: Creator pointing at the text.	Top: MCU, handheld feel. Bottom: MCU, static.	Top: Warm, natural bedroom light. Bottom: Flat studio lighting.	Deliver the final proof of concept and execute the hard Call-To-Action to drive comments and algorithm engagement.

3. Why It Went Viral (Mechanism Breakdown)

The Topic & Psychology

This video taps directly into the current zeitgeist of AI anxiety and excitement. By showcasing an AI avatar that is nearly indistinguishable from a real UGC creator—complete with natural micro-expressions, casual speech patterns, and the ability to hold branded physical products—it triggers a biological "uncanny valley" response that forces viewers to stop scrolling and inspect the footage closely. The target audience includes marketers, indie creators, and tech enthusiasts who are constantly looking for leverage. Psychologically, it leverages the "curiosity gap" by showing an impossible result (AI holding a specific product flawlessly) and immediately promising to reveal the secret workflow. The inclusion of recognizable brands like Colgate grounds the AI in reality, making the technology feel immediately applicable for commercial use, which spikes save and share rates among professionals.

Platform Signals & Algorithm

From a platform perspective, this video is engineered for maximum algorithmic distribution. The split-screen format is a proven retention hack; it gives the eye two places to look, preventing the viewer from swiping away during the first 3 seconds. The pacing is relentless, with visual changes or UI zooms happening every few seconds. However, the true viral engine is the ManyChat automation loop. By gating the actual step-by-step PDF behind a "Comment AI" CTA, the creator artificially inflates the comment count (over 13,000). The algorithm sees this massive influx of comments as a signal of extreme relevance and pushes the video to a broader audience, creating a self-sustaining viral loop.

5 Testable Viral Hypotheses

The Split-Screen Retention Hack: Evidence: The persistent top/bottom layout. Mechanism: Cognitive load is slightly increased, forcing the viewer to process two streams of information, which delays the decision to scroll past. Replication: Use a reaction or "show and tell" split-screen format for tutorial content rather than just a full-screen screen recording.
The "Familiar Object" Reality Anchor: Evidence: The AI holding a recognizable Colgate box. Mechanism: AI often struggles with text and specific objects. Showing it succeed with a known brand breaks the viewer's assumption of AI limits, causing shock. Replication: When showcasing AI generation, include a specific, branded, or highly detailed physical prop to prove the model's fidelity.
The Automated Engagement Loop: Evidence: The "Comment AI" text and the resulting 13k comments. Mechanism: Gating high-value resources behind a low-friction action (typing two letters) trains the algorithm that the content is highly engaging. Replication: Never put the link in the bio; always use an auto-DM tool (like ManyChat) triggered by a specific keyword in the comments.
The "UGC Aesthetic" Camouflage: Evidence: The AI woman's messy bun, casual clothing, and bedroom background. Mechanism: It bypasses the viewer's "ad blocker" instinct because it looks exactly like organic TikTok/Reels content, not a polished corporate ad. Replication: Prompt AI video generators for "iPhone front-facing camera, messy bedroom, casual clothing, natural lighting" rather than "cinematic 4k."
The "Proof of Workflow" Trust Builder: Evidence: Flashing the UI of Arcads and the image editor. Mechanism: Claims of "easy AI" are met with skepticism. Showing the actual interface, even briefly, validates the claim and encourages saves for later reference. Replication: Don't just show the result; include 3-5 seconds of high-speed screen recording showing the actual tool interface being used.

4. How to Recreate (From 0 to 1)

Step 1: Define the Persona & Scene. Decide on your AI creator's look. For this UGC style, aim for relatable and casual. Prompt example: "Young woman, 20s, messy brown hair, wearing a casual red long-sleeve top, standing in a cozy bedroom with bookshelves and a bed in the background, natural window lighting."
Step 2: Generate the Base Image. Use a platform like Arcads (as shown) or Midjourney to generate the initial high-quality image of your avatar holding a generic object (like a coffee cup). Ensure the hands look realistic.
Step 3: Prop Swapping (Inpainting). Take the base image into an AI image editor with inpainting capabilities (like Photoshop Generative Fill or the tool shown in the video). Mask the coffee cup and prompt for the specific product you want (e.g., "a small plush labubu doll" or "a box of Colgate toothpaste"). Generate two versions: one with the product down, one with it raised.
Step 4: Image-to-Video Generation. Use a video model like Kling, Luma Dream Machine, or Runway Gen-3. Use the "Start Frame" and "End Frame" feature. Input your first image (product down) as the start, and the second image (product up) as the end frame. This forces the AI to animate the transition smoothly while maintaining character consistency.
Step 5: Scripting the UGC Hook. Write a script that sounds natural, using filler words ("like," "um") and enthusiastic pacing. Example: "Oh my god guys, do you ever just stop and realize how far AI has actually come?"
Step 6: Audio & Lip Sync. Generate the voiceover using ElevenLabs (choose a conversational, expressive voice). Use a lip-sync tool (like Hedra or SyncLabs) to match the generated audio to your video clip.
Step 7: The Split-Screen Edit. Film yourself reacting to the video. In CapCut or Premiere, stack the videos. Place the AI video on top and your reaction on the bottom. Add a dividing line if necessary.
Step 8: Add the CTA & Publish. Add bold, clear text overlays (e.g., "Comment AI"). Set up a ManyChat flow connected to your Instagram/TikTok to automatically DM the resource to anyone who comments the keyword.

5. Growth Playbook

3 Ready-to-Use Opening Hooks

"I just found a way to make freakishly realistic AI ads, and it's actually terrifying."
"Stop paying for UGC creators until you see what this new AI tool can do."
"This video is 100% AI generated, and I'm going to show you the exact workflow."

4 Caption Templates

The Workflow Reveal: "AI is getting too realistic 🤯 I built this entire UGC ad using just 3 tools. Want the exact step-by-step breakdown? Comment 'WORKFLOW' and I'll send you the free guide. 👇 Have you tried AI video yet?"
The Cost Saver: "Why pay thousands for UGC when AI can do this? 💸 This avatar is completely generated and holds real products perfectly. Comment 'AI' for the tool list! What product would you use this for?"
The Tech Flex: "Sora who? The new image-to-video models are insane. 🤯 Look at how consistent the hands and products are. I put together a PDF of the prompts I used. Comment 'PROMPT' to get it. 👇"
The Skeptic Challenger: "Bet you couldn't tell the top video was AI. 🤖 The lighting, the micro-expressions... it's all generated. Want to learn how to spot it (or make it)? Comment 'GUIDE' and I'll DM you. 👇"

Hashtag Strategy

Broad (Algorithm categorization): #AIVideo #ContentCreation #DigitalMarketing #UGC (These tell the platform what the general topic is, ensuring it reaches people interested in tech and marketing).
Mid-Tier (Targeted communities): #AIHacks #CreatorTips #VideoEditingTutorial #MarketingTools (These target creators and marketers actively looking for new workflows and tools to improve their output).
Niche Long-Tail (Search intent): #ArcadsTutorial #AIAvatarGenerator #ImageToVideoAI #UGCAIGenerator (These capture users specifically searching for the tools mentioned or the exact technique demonstrated, driving high-intent views).

6. FAQ

What tools make it look the most similar?

Based on the video, Arcads is used for the base avatar, an inpainting tool for prop swapping, and models like Kling or Veo for the image-to-video animation.

How do they get the AI to hold the product perfectly?

They don't prompt the video model directly; they generate a still image, use AI inpainting to place the product perfectly in the hand, and then animate that image using start/end frames.

Why does the generated face look consistent throughout the video?

Using the exact same start and end frame images anchors the video generation model, forcing it to maintain the character's identity and clothing between those two points.

How can I avoid making it look like AI?

Prompt for "iPhone front-facing camera," "messy background," and "natural window lighting," and ensure the voiceover includes natural pauses and filler words.

Is it easier to go viral on Instagram or TikTok with this type of content?

This specific "Comment for a DM" strategy works exceptionally well on Instagram Reels due to robust ManyChat integration, whereas TikTok relies more on link-in-bio conversions.

What are the 3 most important words in the prompt?

"UGC style," "natural lighting," and "smartphone camera" are crucial for achieving this specific relatable aesthetic.