0:00 / 0:00

Comment GUIDE and I’ll send it over👇 Stop asking AI for “realistic”. That’s how you get smooth skin, perfect lighting, and plastic faces. Instead, tell the model how the photo was taken. Handheld. Imperfect. Real. I break down the full prompt structure and the final skin realism step in the guide. #realisticai #aiimages #aivideo #aiphotorealistic #aiskin

Gunhild Johanne Reumert

@shedoesai · Digital creator

INSTAGRAM · 2026-01-25Source

163likes

523comments

Remix This

Recreate with Kling 3

Make your own AI viral video

Prompt

GLOBAL LOCK:
Subject: A Caucasian woman in her late 20s, blonde hair tied in a neat ponytail, wearing a leopard-print (cheetah pattern) blouse.
Environment: A cozy home studio/office background with dark grey walls, wooden bookshelves filled with books, green indoor plants, and soft dual-tone lighting (warm orange light from one side, cool blue light from the other).
Camera: MCU (Medium Close-Up) framing, eye-level, 35mm lens feel with shallow depth of field.
Style: Professional UGC creator aesthetic, high-quality video, crisp audio.
Speech: Direct-to-camera delivery, energetic and authoritative tone.

[00:00–00:05]
Visual: Rapid montage of extreme macro close-ups (ECU). First, a human eye with visible iris patterns and eyelashes. Second, an ear with a gold hoop earring showing skin texture. Third, a wrist with a simple black line tattoo showing skin pores and fine hairs.
Action: Static macro shots.
Lighting: Bright, natural daylight feel for the macros.
Text Overlay: "most AI" -> "look fake" -> "because" -> "is trained".
Speech: "Most AI images look fake for one reason. Because AI is trained to remove flaws."

[00:05–00:11]
Visual: The woman (Subject) in the MCU studio setting, gesturing with her hands. Floating icons of AI tools (ChatGPT, Freepik, Ideogram, Nano Banana) appear around her.
Action: Subject talks directly to the camera, moving hands to emphasize points.
Lighting: Studio setup (Orange/Blue).
Text Overlay: "need" -> "AI tools" -> "to prompt".
Speech: "But we don't need better AI tools. We just need to prompt the model to create images that actually look real."

[00:11–00:21]
Visual: Transition to a black screen with white text titled "Master Prompt". The text scrolls or highlights specific sections. Then, a split screen showing the woman talking in a small window and the prompt text in a larger window.
Action: Subject continues talking while the prompt text is displayed.
Lighting: Studio setup for the talking head.
Text Overlay: "to create" -> "that actually" -> "look real".
Speech: "The key to realistic AI images is using a prompt with a specific structure. This prompt should force skin detail, including visible pores, uneven tone, and natural imperfections."

[00:21–00:30]
Visual: Montage of AI-generated faces with high realism. A man's face with stubble and pores, a woman's face with freckles and slight redness. Then, a screen recording of the Freepik interface showing a gallery of realistic portraits.
Action: Fast cuts between the portraits and the UI.
Lighting: Varied, matching the generated images.
Text Overlay: "most people start" -> "make" -> "image".
Speech: "Most people start their prompt with 'make a realistic image of'. I start by telling the model how the camera behaves."

[00:30–00:42]
Visual: Screen recording of a prompt being typed into a text box. Keywords like "iPhone 14 Pro", "handheld framing", and "imperfect composition" are highlighted in yellow.
Action: Scrolling through the prompt text.
Lighting: Digital UI.
Text Overlay: "model that" -> "camera behaves" -> "casual hand" -> "imperfect composition".
Speech: "Casual handheld framing, slightly imperfect composition, and a smartphone camera perspective. This alone already breaks the AI look."

[00:42–00:52]
Visual: The woman back in the MCU studio setting. She gestures toward floating app icons for "Enhancor" and "Higsfield". A screen recording shows a "Skin Enhancer" tool being used on a photo of a woman with goggles.
Action: Subject explains the final step.
Lighting: Studio setup.
Text Overlay: "But Most People Stop There" -> "Final Step" -> "Most Creators Are Gatekeeping".
Speech: "But most people stop there. I use a final step that most creators are gatekeeping. I run each image through a final skin enhancement step using Enhancor or Higsfield."

[00:52–01:00]
Visual: The woman in MCU, pointing down toward a text box that says "Comment GUIDE". A final zoom-out effect or a slight blur transition.
Action: Subject smiles and points.
Lighting: Studio setup.
Text Overlay: "Prompt Structure" -> "Workflow" -> "Comment GUIDE".
Speech: "If you want my exact prompt structure and the full workflow, just comment GUIDE and I'll send it over."

NEGATIVE PROMPT:
Smooth skin, plastic texture, perfect symmetry, airbrushed look, 6 fingers, distorted eyes, watermark, logo, blurry background (unless specified), robotic voice, lip-sync lag, harsh sibilance, flickering lights, low resolution.

SPEECH PACK:
[00:00-00:05] "Most AI images look fake for one reason. Because AI is trained to remove flaws."
[00:05-00:11] "But we don't need better AI tools. We just need to prompt the model to create images that actually look real."
[00:11-00:21] "The key to realistic AI images is using a prompt with a specific structure. This prompt should force skin detail, including visible pores, uneven tone, and natural imperfections."
[00:21-00:30] "Most people start their prompt with 'make a realistic image of'. I start by telling the model how the camera behaves."
[00:30-00:42] "Casual handheld framing, slightly imperfect composition, and a smartphone camera perspective. This alone already breaks the AI look."
[00:42-00:52] "But most people stop there. I use a final step that most creators are gatekeeping. I run each image through a final skin enhancement step."
[00:52-01:00] "If you want my exact prompt structure and the full workflow, just comment GUIDE and I'll send it over."

How shedoesai Made This Skin Texture Prompting AI Video

This case study analyzes a high-performing educational reel by @shedoesai that deconstructs the "uncanny valley" of AI-generated images. The video effectively uses a cinematic editorial portrait aesthetic combined with UGC-style direct-to-camera delivery. By contrasting "plastic" AI faces with extreme macro shots of human imperfections (freckles, pores, uneven skin tones), the creator establishes immediate authority. The core value proposition is a shift from generic prompting to technical "camera-first" descriptions, utilizing tools like Google Nano Banana Pro and Enhancor. This strategy targets indie creators looking to move beyond the "AI look" toward professional-grade realism.

What You’re Seeing

The video is a fast-paced montage that blends high-quality talking-head footage with extreme macro photography and screen recordings. The creator is positioned in a home studio with dual-tone practical lighting (warm orange and cool blue) which creates depth and a professional "tech-creator" vibe. She wears a leopard-print shirt, providing a busy texture that contrasts well with the clean, blurred background of bookshelves and plants.

Visually, the video relies on extreme close-ups (ECU) to prove its point. We see the fine details of a human eye, the texture of an earlobe with a gold hoop, and the subtle redness of skin. These are interspersed with screen recordings of the Freepik interface and text overlays that highlight specific prompt keywords like "handheld," "imperfect," and "smartphone camera perspective." The editing rhythm is snappy, with cuts occurring every 1.5 to 2 seconds to maintain high retention.

Shot-by-Shot Breakdown

Time Range	Visual Content	Shot Language	Lighting & Tone	Viewer Intent
00:00–00:05	Extreme macro of eye, ear, and skin texture.	ECU / Static	Natural, high detail	Hook: Challenge the "AI look" with real texture.
00:05–00:12	Creator talking with floating AI tool icons.	MCU / Eye-level	Studio (Orange/Blue)	Establish authority and define the problem.
00:12–00:22	Text overlay of "Master Prompt" on black background.	Graphic / Text	High contrast	Tutorial value: Provide the "secret sauce."
00:22–00:35	Montage of realistic AI faces vs. real skin details.	Fast cuts / ECU	Varied (Realism focus)	Reinforce the "Imperfect = Real" persona.
00:35–00:50	Screen recording of Freepik and Skin Enhancer tools.	UI Capture	Clean / Digital	Actionable steps: Show the exact tools used.
00:50–01:00	Creator recap with "Comment GUIDE" overlay.	MCU / Eye-level	Studio (Orange/Blue)	CTA: Convert viewers into leads.

Why It Went Viral

The success of this video lies in its counter-intuitive hook. Most AI tutorials promise "perfect" results; this video does the opposite by telling users to stop asking for "realistic" and start asking for "imperfect." This taps into the psychological desire for authenticity in an increasingly synthetic digital landscape. By framing "smooth skin" and "perfect lighting" as failures, the creator creates a "pattern interrupt" that stops the scroll.

Furthermore, the video leverages high utility and "gatekeeping" psychology. The creator explicitly mentions that other creators are "gatekeeping" these steps, positioning herself as the transparent expert. This builds massive trust and encourages "Saves" because the information (the prompt structure) is too dense to memorize in one viewing.

Platform Perspective: Instagram's algorithm prioritizes "Watch Time" and "Saves" for educational content. The rapid-fire visual evidence of skin texture ensures users don't look away, while the promise of a "Guide" in the comments drives high engagement (comments) and signals to the algorithm that the content is highly valuable, triggering a wider distribution to the Explore page.

5 Testable Viral Hypotheses

The "Anti-Keyword" Hook: Telling users not to use a popular keyword (e.g., "realistic") creates a curiosity gap that increases 3-second retention.
Macro-Visual Proof: Using extreme close-ups of "flaws" (pores, veins) acts as a biological hook, as the human eye is naturally drawn to high-detail organic textures.
The "Tool Stack" Reveal: Mentioning niche, high-performance models like "Google Nano Banana" creates a sense of "insider knowledge" that drives shares.
Contrast Pacing: Alternating between a talking head and high-speed UI recordings prevents "visual fatigue" and keeps the viewer engaged for the full 60 seconds.
The "Comment for Guide" Loop: Using an automated DM trigger (like ManyChat) converts passive viewers into active commenters, exponentially increasing the post's reach.

How to Recreate (Step-by-Step)

Identify the "Uncanny Valley": Find a common AI mistake (e.g., plastic skin, 6 fingers) and position your video as the solution to "fixing" it.
Set the Scene: Use a 35mm or 50mm lens feel for your talking head. Use practical lights (LED bars) in the background to create a "pro-creator" depth of field.
Capture Macro B-Roll: If you can't film your own skin, use high-quality stock footage of extreme close-ups showing freckles, pores, and natural imperfections.
Draft the "Master Prompt": Create a structured prompt that focuses on camera behavior (e.g., "shot on iPhone 14 Pro," "handheld," "slight motion blur") rather than just subject description.
Highlight Key Phrases: In your edit, use bold text overlays to highlight the specific words that "break" the AI look (e.g., "Uneven Tone", "Visible Pores").
Show the Workflow: Record your screen while using a specific tool (like Freepik or Flux). Don't just talk about it; show the "Generate" button being clicked.
Add a Post-Processing Step: Introduce a "secret" final step, like a skin enhancer or a grain filter, to add that final layer of realism.
Deploy the CTA: End with a clear instruction to comment a specific word to get the "Full Guide."

Growth Playbook

3 Opening Hook Lines

"Stop using the word 'realistic' in your AI prompts. Here's why it's ruining your images."
"The secret to making AI look 100% real isn't the model, it's the camera behavior."
"I found the prompt structure that most AI creators are gatekeeping."

4 Caption Templates

Template 1 (The Problem/Solution): Stop asking AI for "perfect." ❌ That’s how you get plastic faces and fake lighting. Instead, tell the model how the photo was taken. Handheld. Imperfect. Real. I’m sharing my full Master Prompt structure today. Comment "REAL" and I'll DM it to you! 👇

Template 2 (The Tool Reveal): Most people use generic models. I use Google Nano Banana Pro. 🍌 The difference in skin texture is insane. See the pores? The uneven tones? That’s the secret. Want the full workflow? Comment "GUIDE" below! 🚀

Hashtag Strategy

Broad (Reach): #AIArt #GenerativeAI #ArtificialIntelligence #ContentCreator
Mid-Tier (Niche): #AIPrompting #DigitalArtTips #AITutorial #FreepikAI
Long-Tail (Conversion): #RealisticAI #SkinTextureAI #IndieCreatorTools #AIGrowthHacks

FAQ

What tools make AI images look the most similar to real photos?

Using models like Flux.1 or Google Nano Banana Pro combined with skin-specific enhancers like Enhancor provides the best results.

What are the 3 most important words in a realism prompt?

"Handheld," "Imperfect," and "Smartphone-perspective" are crucial for breaking the studio-perfect AI look.

Why does the generated face look inconsistent?

AI models often default to "beauty" standards; you must explicitly prompt for "asymmetrical features" and "natural skin texture" to maintain consistency.

How can I avoid making it look like AI?

Avoid high-key studio lighting prompts and instead use "natural ambient light" or "slight underexposure."

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

Instagram is currently better for high-aesthetic "Save-able" educational content, while TikTok favors raw, unpolished "behind-the-scenes" reveals.