Red’s Run
Vector City, 2144.
Red, a 5ft-tall mercenary courier, takes on a shadowy job: deliver a cryptic package to The Grandmother, the enigmatic crime lord who rules the neon-blackened underworld.
But ruthless rival gangster The Wolf has other plans, setting off a deadly hunt through the grimy streets of this oppressive urban jungle.
And somewhere behind it all, The Hunter moves silently, tracking his mark.
In this neo-noir reimagining of a classic fable, Red descends into a brutal game of predator and prey, where assassins, corrupt AIs, and back-alley betrayals lurk at every turn. One wrong move and her armored hyperbike might end up shredded metal in the gutters of the city’s unforgiving expanse.
Ok, so you’re the kind of visual nerd who reads the footnotes. Respect.
My workflow for Red’s Run started by mapping out the scenes I wanted, then building the story in stills.
Each shot was treated like a screencap from a real sci-fi film, composed as if it were a storyboard frame, but with cinematic lighting, camera placement, and atmosphere already baked in.
I created a prompt system to maintain character consistency across different environments and angles. For that, I used a set of custom GPTs and Gemini Gems that expand basic concepts into ultra-detailed, prompt-optimized descriptions.
You can try one of the ones I built HERE.
Final design pass for The Wolf
Mask design consistency across scenes
Most of the base imagery was generated using Google Imagen 3, which, as of the publishing of this post, has in my opinion the best prompt adherence + cinematic look combo of any frontier model.
After plenty of re-rolls, I took the selected stills into Krea AI Realtime, MidJourney Editor, and good ol’ Photoshop to build visual continuity, especially around backgrounds, facial structure, and costume detail.
Once the stills were locked, I manually refined each one: comping additional layers and characters, recoloring, fixing lighting inconsistencies, and designing additional elements like jacket badges, weapons, and environmental signage.
Step-by-step build of the rooftop shot, from base plate to final comp
The final scene in motion
Early character variations for the guards and prisoner lineup
For worldbuilding details, every badge on Red’s jacket was designed as a nod to the last 50 years of cyberpunk. The logos reference fictional corporations from across the genre. Including deep cuts like Shadowrun 2E and Hardwired. The sniper’s HUD for example was built from scratch.
Original cyberpunk corp logos, designed as in-world jacket patches
A mini-library of sniper HUD elements created from scratch for the crosshairs sequence
Just a background detail, until you zoom in and read it
When everything was in place, I ran multi-pass upscaling using Magnific AI and Topaz to push the visual fidelity as far as I could.
I’ve found ultra-high-resolution stills to be essential when moving into image-to-video generation. Video models tend to compress, blur, or soften detail, so starting from a strong foundation matters.
The final stills were then run through three different models: Runway Gen-4, Kling 2.0, and Google Veo 2. For this step, I used additional custom GPTs trained to optimize prompts for video generation.
After many re-rolls (and many credits), I locked in my favorite shots, upscaled everything to 4K, and edited it all on CapCut Desktop… because who needs Premiere in 2025? 🤷
Voiceover was generated using Eleven Labs. Sound design by yours truly, paired with a custom music edit of a stock track.
And voilà: from imagination to moving image.