Overview
Palms Motel is a fictional 1970s Palm Springs motel that never existed — told entirely through AI-generated images on TikTok. Each post was a slideshow: six photographs of pools, interiors, and sun-bleached architecture, captioned POV: you just moved to Palm Springs, 1977, set to a viral audio track.
The project ran for 48 posts. One hit 770,000 views and 105,000 likes. Total likes across the account reached 109k. A Shopify store sold poster prints of the images. The whole thing was built on a single engineered system — a prompt framework designed to produce visually consistent, series-quality images at scale.
The Idea
The premise was simple: build a world, not just an image. TikTok's slideshow format meant viewers weren't just seeing a single AI image — they were moving through a space, room by room, from pool to lobby to courtyard. The POV caption put them inside it.
What made it work wasn't the individual images. It was the consistency. Every post felt like it was photographed on the same afternoon, in the same light, by the same person. That coherence is what made the account feel like a real place rather than a collection of AI experiments.
Achieving that consistency at scale required building a system.
The Prompt Engineering System
Midjourney's outputs are inherently variable. Without a deliberate framework, a series of 48 posts across dozens of AI generations would drift — colour temperatures shifting, rendering styles changing, the aesthetic falling apart post by post.
The solution was to treat prompt writing as a design discipline. I developed a consistency framework that separated every prompt into fixed and variable components.
Fixed elements — never changed across the series:
Era and publication reference:
"Photograph from a 1978 lifestyle magazine"
Camera and lens specification:
"UHD, 24k Hasselblad x1 camera, 28mm lens"
Colour temperature:
A specific sage-coral combination held across every image
Location:
Palm Springs, California or similar
Photographer reference:
Slim Aarons, staged for Architectural Digest
Parameters:
`--no filter texture overlay dirt grain film filter --ar 4:3 --stylize 800 --weird 45`
Variable elements — changed for each post:
Primary subject:
pool shape, interior room, architectural feature
Spatial arrangement:
courtyard, colonnade, grove, sanctuary
Prominent focal point:
what draws the eye in this specific frame
This structure meant every generation stayed inside the same aesthetic world while still producing images that were distinct from one another. The fixed elements did the work of cohesion. The variable elements did the work of interest.
LOW-ANGLE, BUNGALOW, KIDNEY BEAN SHAPE POOL, TOWELS, CLOSE UP, MIDCENTURY MODERN, ART DECO, CACTUS, RETRO, 1983, STAGED FOR ARCHITECTURAL DIGEST MAGAZINE TAKEN WITH HASSELBLAD XD1, 400MM LENS, SLIM AARONS --NO FILTER --RAW --STYLIZE 250
HUGH KAPTUR WHITE BUNGALOW, KIDNEY BEAN SHAPED POOL, CLOSE UP, MIDCENTURY MODERN, ART DECO, CACTUS, RETRO, PINK OPUNTIAS FLOWERS, 1983, STAGED FOR ARCHITECTURAL DIGEST MAGAZINE TAKEN WITH HASSELBLADYDI, 400MM LENS SLIM AARONS --NO FILTER --RAW --STYLIZE 250
Turning the System into a Tool
To maintain consistency across 48 posts — and to scale generation without rewriting prompts from scratch each time — I documented the framework into a prompt guide and used it as a briefing document for AI-assisted prompt generation.
The guide defined:
- The aesthetic DNA of the series (era, colour, light, materiality)
- The fixed technical parameters and why they existed
- A prompt template with clearly labelled placeholders for variable elements
- A quality control checklist to validate each generation before it was used
Feeding this guide to an AI allowed me to generate batches of new, on-brand prompts quickly — the system handled consistency while I made the creative judgement calls about which outputs were worth keeping.
Production & Post-Processing
Raw Midjourney outputs went through Photoshop before they were used — colour grading to lock the warmth, occasional compositing to fix AI artefacts, and cropping and export optimised for TikTok's slideshow format.
Each post was assembled as a six-image sequence: a loose spatial narrative moving through the space, with enough variety in framing and subject to hold the viewer through the whole slideshow. Audio selection was deliberate — tracks chosen for nostalgia and existing viral traction to increase the chance of algorithmic pick-up.
The Store
The strongest images were adapted as poster prints and listed on a Shopify store — cropped, colour-corrected, and exported at print resolution in a clean poster format. The store ran briefly before being closed: the time cost of maintaining both the content cadence and the store outweighed the revenue at that stage of the project.
Results
48
Total Posts
288+
Images Generated
770,000
Total Post Views
105,000
Top post likes