From a simple viral video to a military-style action comedy featuring me, my dog, and one enormous tank — created with Higgsfield AI.
Sometimes the best AI video ideas start with something incredibly simple.
I came across a viral clip featuring a woman casually sitting in a folding chair while a large Toyota vehicle pulled up behind her, surrounded by a spectacular cloud of dust.
The scene was dramatic, but what made it funny was how completely unfazed she looked.
That contrast gave me an idea.
What if I replaced the woman with myself, swapped the Toyota for a massive military tank, and put my 15-pound Toy Aussiedoodle, Ollie, in command?
That became the latest installment of my ongoing AI video series: Full Metal Ollie.
The final result was a cinematic, roughly 13-second video featuring an enormous tracked vehicle emerging through clouds of dust, with Ollie commanding the tank while I sit in the foreground as though nothing unusual is happening.
Here's how I developed it using Higgsfield, reference images, and a little prompt engineering.
Step 1: Understanding Why the Original Video Worked
Before generating anything, I studied the composition of the original viral clip.
There were three elements I wanted to preserve:
- The stationary camera: Keeping the viewpoint fixed makes the approaching vehicle feel enormous.
- The dramatic entrance: The combination of vehicle movement, sunlight, and swirling dust creates a cinematic reveal.
- The human reaction: The person in the chair remains completely calm despite the spectacle unfolding behind them.
I realized the humor wasn't necessarily in the vehicle itself. It was the contrast between an ordinary person sitting casually and something extraordinary happening in the background.
I wanted to reproduce that formula while introducing my own characters and visual style.
Step 2: Building Consistent Characters With Reference Sheets
One of the biggest challenges in AI video generation is character consistency.
Without good references, AI models can alter facial features, clothing, body proportions, and even the number of limbs from frame to frame.
For this project, I had two important character references.
Christopher Avalos — My AI Character Sheet
I created a character sheet using my facial references, including front, profile, and three-quarter views.
The sheet also established my clothing and general appearance: a navy blazer, white open-collar shirt, dark jeans, and brown shoes.
This helped define the character I wanted sitting in the folding chair.
The important detail was separating genuine facial references from outfit and body references. I wanted the AI to reproduce my appearance rather than invent someone who merely looked similar.
Full Metal Ollie — The Tank Commander
Ollie is my real Toy Aussiedoodle and the star of several AI-generated adventures.
For this video, I used our previously developed Full Metal Ollie character and military tank references.
The design features Ollie wearing a vintage camouflage helmet and military uniform, emerging from the commander's hatch of an enormous weathered olive-green tank.
His recognizable facial markings, fluffy black coat, tan eyebrows, white chest, floppy ears, and expressive brown eyes were essential to maintaining his identity.
I also had reference views of the tank from several angles to establish its shape, scale, tracks, turret, and overall appearance.
Creator tip: A good character sheet is an investment. Once you have established a character's appearance, you can reuse that reference across many different scenes rather than redesigning the character every time.
Step 3: My First Prompt Was Too Complicated
My first attempt was ambitious.
I described the entire scene in detail, including the characters, military vehicle, dust effects, camera placement, timing, and dramatic action.
I even broke down the sequence second by second.
The prompt was cinematic and technically descriptive.
Unfortunately, Higgsfield rejected the generation with an NSFW safety warning.
The credits were refunded.
I revised the prompt, making the scene more obviously humorous and removing much of the military action language.
I submitted it again.
Another NSFW rejection. Another credit refund.
I couldn't determine precisely what was triggering the system. It may have involved the combination of military imagery, character references, or other elements in the request.
But it taught me an important lesson:
More detailed prompts are not automatically better prompts.
Sometimes a generation model doesn't need an elaborate screenplay. It needs a clear description of what should be visible and how the subjects should move.
Step 4: Simplifying the Prompt
Rather than continually expanding and modifying the original prompt, I returned to the basic concept.
I stripped the description down to the essential visual ingredients: a man sitting in a chair, a large tracked vehicle approaching behind him, Ollie looking out from the hatch, and dramatic dust illuminated by golden sunlight.
Here is the simplified prompt that became the basis of the successful generation:
The Higgsfield Prompt
Cinematic 9:16 video, one continuous stationary camera shot.
A friendly middle-aged man with dark hair and a short salt-and-pepper beard sits comfortably in a black folding camping chair on a gravel clearing.
Behind him, a large vintage olive-green tracked vehicle slowly rolls into view and stops. A small black-and-tan fluffy dog named Ollie, wearing a camouflage outfit and helmet, looks out from an open roof hatch.
Soft golden sunlight illuminates a dramatic cloud of dust rising behind the vehicle.
The man remains perfectly calm and smiles slightly at the camera. Ollie tilts his head playfully.
Photorealistic cinematic comedy, realistic proportions, natural movement, consistent characters, warm desert atmosphere, no dialogue.
Notice what's missing.
There are no complicated camera instructions, lengthy choreography descriptions, or elaborate effects breakdowns.
The entire scene is communicated through a handful of clearly defined visual instructions.
That was a much better fit for the concept I was trying to create.
Step 5: The Successful Generation
The resulting video came together beautifully.
The camera remains fixed while the large tracked vehicle appears behind my chair.
Dust fills the background, creating the kind of dramatic atmosphere you'd expect from a Hollywood action scene.
Ollie appears in the hatch, looking like the world's smallest and most confident tank commander.
Meanwhile, I'm sitting comfortably in the foreground, apparently unconcerned that an enormous military vehicle has just arrived behind me.
The final result was approximately 13 seconds long.
And the best part?
The entire joke works without a single line of dialogue.
The visual contrast tells the story.
The heavy vehicle, cinematic dust, tiny dog commander, and casual seated human create the comedy.
Step 6: What I'd Do Differently Next Time
This experiment reinforced several important lessons for my future AI video projects.
Start with the composition, not the effects. The original viral video's strength was its visual arrangement. Preserving that arrangement mattered more than adding complicated action.
Use consistent character references. This is especially important when working with recurring characters such as Ollie, whose recognizable appearance is central to the series.
Keep prompts focused. The successful version was significantly simpler than the original attempts. It gave the model room to interpret natural movement without losing the overall concept.
Treat safety rejections as troubleshooting information. A rejected generation doesn't necessarily mean the creative idea is impossible. The prompt, reference images, and selected model can all influence the outcome.
Test before committing to expensive renders. When available, lower-resolution previews are useful for checking character consistency, movement, anatomy, and overall composition before spending more credits on a final version.
One especially useful future technique is creating a single composite starting image featuring all the characters and vehicle already positioned correctly. That could reduce the number of separate visual elements the video model must reconcile.
Step 7: Turning One Experiment Into a Recurring Character Series
This wasn't my first Full Metal Ollie concept, and it won't be the last.
What started as experimenting with AI-generated images of my dog has gradually evolved into a collection of cinematic scenes, character references, fictional trailers, and short-form comedy videos.
I'm particularly interested in combining recognizable internet video formats with original characters and humorous storytelling.
Instead of simply reproducing a viral clip, I want to transform its central idea into something new.
In this case, a dusty car entrance became a military-style movie moment starring a tiny Aussiedoodle.
That transformation is one of the things I enjoy most about generative AI.
Final Thoughts: Simple Ideas, Big Results
Creating memorable AI video content doesn't always require a complicated storyline, elaborate dialogue, or dozens of visual effects.
Sometimes it starts with recognizing why a simple scene is entertaining.
The creative process is about identifying that essential ingredient, adding your own perspective, and experimenting until everything comes together.
For Full Metal Ollie, that meant a folding chair, an enormous tank, some cinematic dust, and one very confident little dog.
Mission accomplished.
Create Your Own With Higgsfield
If you're interested in experimenting with cinematic AI videos, character references, or original short-form content, you can explore the tools I've been using at Higgsfield.
Try Higgsfield AI — My Affiliate Link
Affiliate disclosure: This is my referral link. I may receive a commission or referral benefit if you sign up or make a qualifying purchase.
Follow Our AI Adventures on Instagram
Want to see more cinematic experiments, creative AI workflows, and behind-the-scenes projects?
Follow my AI filmmaking experiments, visual storytelling, and creative projects.
Ollie's Instagram — @toy.aussie.ollie
Follow Ollie's adventures, hilarious AI-generated videos, and his ongoing career as the world's smallest action hero.
More Full Metal Ollie adventures are coming!
Created by Christopher Avalos | Evilos AI Studios
Tags: Higgsfield AIVideo FullMetalOllie AIFilmmaking PromptEngineering CreativeAI ViralVideo CharacterConsistency
