0:00 / 0:00

onofumi.ai

@onofumi.ai

INSTAGRAM · 2025-12-06Source

931likes

9comments

Remix This

Recreate with Wan2.6

Make your own AI viral video

Prompt

GLOBAL LOCK: The video features a series of anthropomorphic fruits with hyper-realistic, grotesque human facial features. The subjects include a red apple, a yellow lemon, and an orange. All fruits have deeply wrinkled skin, realistic human teeth, and expressive eyes. Tiny 1:87 scale miniature construction workers wearing orange high-visibility vests and yellow hard hats are positioned on and around the fruits, performing labor with shovels and scaffolding. The setting is a weathered wooden surface, likely a deck or table, with a soft-focus green foliage background. The lighting is natural, diffused daylight. The camera uses a macro lens aesthetic with a very shallow depth of field. The color grade is cinematic with high texture contrast and natural saturation. Audio consists of guttural, distorted vocalizations and wet, crunchy ASMR sound effects.

[00:00–00:04]
A close-up of a red apple with a grumpy, wrinkled human face. Its mouth is slightly open, revealing realistic teeth and gums. Several miniature workers are on top and at the base; one worker on the left is shoveling the apple's skin. The apple's face contorts as if groaning, with the jaw moving up and down. The camera is static, focusing on the facial movement. Guttural "Grah!" sound effect synced to the jaw opening.

[00:04–00:08]
Transition to a yellow lemon with a wide, laughing or screaming human face. The skin is bumpy and textured. Miniature workers are on top, and a small metal scaffolding structure is built on the right side with a worker standing on it. The lemon's mouth opens wide in a silent shout, tongue visible. The camera angle is slightly wider. High-pitched, distorted "Yeah!" vocalization.

[00:08–00:12]
Transition to an orange with an extremely angry expression, bulging eyes, and a furrowed brow. Workers are shoveling orange pulp at the base. Suddenly, thick red liquid (resembling blood or lava) erupts from a hole in the top of the orange's head like a volcano. Simultaneously, the orange opens its mouth wide and sticks out a long tongue as red liquid flows out. The eruption is violent with droplets splashing. Sound of a wet explosion and squelching.

NEGATIVE PROMPT: Smooth skin, cartoonish features, low resolution, blurry textures, plastic-looking fruit, static faces, silent audio, bright studio lighting, clean clothes on workers, floating objects, flickering liquid, robotic movement, watermark, text.

SPEECH PACK:
[00:00-00:04]
Transcript: [Guttural groan/shout] "GRAAAH!"
TAKE_A: Deep, gravelly, low-pitched roar.
TAKE_B: Short, sharp, pained grunt.
TAKE_C: Distorted, monster-like vocalization.

[00:04-00:08]
Transcript: [Distorted shout] "YEAAAH!"
TAKE_A: High-energy, manic laughter-scream.
TAKE_B: Aggressive, distorted cheer.
TAKE_C: Glitchy, pitched-up vocal burst.

[00:08-00:12]
Transcript: [Eruption sound + Gurgle]
TAKE_A: Wet, bubbling sound followed by a deep "BLARGH".
TAKE_B: Explosive "POP" followed by liquid flowing sounds.
TAKE_C: Visceral, squelching eruption noise.

How onofumi.ai Made This Grotesque Miniature Fruit Construction AI Video — and How to Recreate It

This viral case study examines a high-impact "Grotesque Miniature" aesthetic created by @onofumi.ai. The video features hyper-realistic, anthropomorphic fruits—apples, lemons, and oranges—sporting intensely expressive, wrinkled human faces. These "fruit-monsters" are being swarmed by tiny, 1:87 scale-style construction workers who appear to be mining or repairing the fruit's flesh. Combining macro-cinematography, uncanny valley psychology, and crunchy ASMR sound design, this format taps into the "oddly satisfying yet disturbing" niche that consistently performs well on Instagram and TikTok. The core appeal lies in the extreme texture detail and the surreal juxtaposition of mundane labor with biological horror.

What You’re Seeing

The video is a series of three distinct macro-dioramas. Each scene features a different fruit as a central "character" reacting to the presence of miniature workers. The lighting is soft, naturalistic daylight, suggesting an outdoor wooden table or deck, which grounds the surreal subject matter in reality. The depth of field is extremely shallow (bokeh), emphasizing the "macro" lens feel common in miniature photography.

Shot-by-Shot Breakdown

Time Range	Visual Content	Shot Language	Lighting & Tone	Viewer Intent
00:00 – 00:04	A red apple with a grumpy, toothy human face. Workers shovel bits of its "skin."	Macro Close-up (MCU), static with internal motion.	Warm, natural light; earthy tones.	The Hook: Immediate "What am I looking at?" factor.
00:04 – 00:08	A yellow lemon/citrus with a laughing/screaming face. Scaffolding is erected around it.	Medium Macro, slightly wider to show scale.	Bright, high-saturation yellow; high energy.	Retention: Escalates the absurdity and introduces complexity (scaffolding).
00:08 – 00:12	An orange with bulging eyes. Red "lava" juice erupts from its head and mouth.	Extreme Close-up (ECU) on the eruption.	High contrast; dramatic "climax" motion.	The Payoff: Visual climax and "loop" setup.

Why It Went Viral: The "Uncanny" Hook

The Psychology of the Grotesque

This content succeeds by leaning into the Uncanny Valley. Humans are biologically programmed to pay attention to faces, especially those that look "off" or highly expressive. By mapping realistic human aging (wrinkles, decaying teeth, wet eyes) onto inanimate fruit, the creator triggers a "disgust-curiosity" loop. Viewers can't look away because their brains are trying to categorize the object: Is it food? Is it a person? Is it a toy?

The "Miniature World" Fascination

There is a long-standing internet subculture dedicated to miniatures (dioramas, tilt-shift photography). This video hijacks that interest but adds a dark twist. The "labor" aspect—tiny workers with shovels and hard hats—adds a narrative layer. It’s not just a weird fruit; it’s a construction site. This narrative depth encourages saves, as users want to show the "creative concept" to others.

Platform Signal Analysis

From a platform perspective, the Watch Time is driven by the density of detail. You cannot see everything in one 12-second loop. You have to watch again to see what the worker on the left is doing, or to look closer at the apple's teeth. The ASMR audio (squishing, crunching, guttural groans) provides a sensory anchor that keeps users from scrolling past, even if they find the visuals slightly repulsive.

5 Testable Viral Hypotheses

The "Texture Overload" Hypothesis: High-frequency detail (wrinkles, pores, fruit skin) triggers longer gaze duration. Replication: Use "hyper-detailed" and "8k macro" in your prompts.
The "Scale Contrast" Hypothesis: Placing something familiar (construction workers) next to something impossible (living fruit) creates instant cognitive friction. Replication: Always include a "human scale" reference point.
The "Visceral Sound" Hypothesis: Wet, squishy sound effects paired with facial movement increase "immersion" and shareability. Replication: Layer 3-4 distinct ASMR sounds per shot.

The "Biological Horror" Hypothesis:

Replication: Don't make it "cute"; make it slightly uncomfortable.

The "Action-Reaction" Hypothesis: Showing the workers "causing" the fruit to scream creates a cause-and-effect narrative that is more engaging than a static image. Replication: Ensure the fruit's expression changes in response to the workers.

How to Recreate: Step-by-Step Tutorial

Step 1: Concept & Fruit Selection

Choose fruits with distinct textures. A strawberry (seeds), a pineapple (scales), or a pomegranate (seeds/juice) work best for AI generation because the textures provide more "data" for the AI to manipulate into facial features.

Step 2: Character Consistency (The Base Image)

Use Midjourney or DALL-E 3 to create your base "Hero Image." Prompt Formula: Macro photography of a [Fruit Type] with a highly detailed, wrinkled human face, expressive [Emotion], tiny 1:87 scale construction workers in orange vests climbing and shoveling the fruit, wooden table background, shallow depth of field, cinematic lighting --ar 9:16

Step 3: Keyframe Generation

Generate two versions of the same scene: one with the mouth closed and one with the mouth open/screaming. This provides the "start" and "end" points for your video AI.

Step 4: Video Generation (The Motion)

Upload your base image to Kling AI, Luma Dream Machine, or Runway Gen-3. Motion Prompt: "The fruit's face screams and grimaces, eyes rolling, while the tiny workers move their shovels up and down. Red juice erupts from the top like a volcano."

Step 5: Sound Design (The Secret Sauce)

Do not use generic background music. Use a site like Epidemic Sound or Freesound.org to find:

Wet squelching sounds
Stone/dirt shoveling sounds
Deep, pitched-down vocal grunts

Step 6: Editing & Pacing

Cut the video every 3-4 seconds. The "switch" to a new fruit resets the viewer's attention span. Ensure the cuts land on a heavy "crunch" or "scream" sound effect.

Growth Playbook

Opening Hook Lines

"Nature is fighting back... 🍎"
"The secret life of your fruit bowl."
"Wait for the orange at the end 🍊🔥"

Caption Templates

The Curiosity Gap: "I’ll never look at an apple the same way again. Which one is your favorite? 👇 #aiart #miniatures #surreal"
The Process Reveal: "Building a tiny world, one bite at a time. 🏗️ Created with [AI Tool Name]. Thoughts on this aesthetic? #creative #aiart #macro"
Short & Punchy: "Fruit Salad: The Horror Movie. 💀 #oddlysatisfying #uncannyvalley"

Hashtag Strategy

Broad (High Volume): #aiart #cgi #3dart #surrealism
Mid-Tier (Niche Interest): #miniatureworld #macro_creativity #oddlysatisfying #horrorart
Long-Tail (Specific): #anthropomorphic #fruitart #aiartist #onofumi_style

Frequently Asked Questions

What tools make it look the most similar?

Midjourney for the base image and Kling AI for the high-fidelity facial motion.

What are the 3 most important words in the prompt?

"Macro," "Wrinkled," and "Anthropomorphic."

Why does the generated face look inconsistent?

Use a "Character Reference" (--cref in Midjourney) to lock the face before animating.

How can I avoid making it look like "bad" AI?

Focus on lighting consistency and high-quality sound design; bad audio ruins good AI video.

Is it easier to go viral on Instagram or TikTok?

Instagram Reels currently favors high-aesthetic, "uncanny" visual loops like this.