AI Logs — reionthewall

REIONTHEWALL

rei is a left + right brain digital polymath / degen living in Brooklyn, NY working at the intersection of technology, culture, and research. hands in (and on) all things interdisciplinary—digital transformation, AI research / applications, strategy, product/ion, UX/UI design and development, blockchain, information & CMS architecture, design thinking, human-centered design, HCI, and applied research. deeply into people and the internet. anthropology + ethnography + literature bg from University of St. Andrews. currently leading MATTE’s digital team as Director, Digital and Enterprise Innovation. interested in digital culture, thinking about things too deeply, building AI art and applications, drawing + reading comics, and all things web2, 2.5, and 3. NFT NYC selected artist ‘23 + 24 korean / minnesotan

SOME OF HUNDREDS OF INTERMITTENT AI LOGS

[2021] Disco Diffusion, Text Soup + Digital Alchemy

Medium: Google Colab, VQGAN + CLIP
Focus: Create and remix surreal environmental scenes
Notes: Remixing visual prompts using vqgan_imagenet_f16_16384 w/ keyword blending. Painting with language—minimal control over textures/worlds but better than nothing. In Disco Diffusion discord. Prompt hacking + collaboration. Prompt engineering = less science, more alchemy. Iteration, randomness, remixing = foundation of AI image generation.

BLUE EYED LUCY [THE SNAKE BUT NOT THE SNAKE]

SCREENSHOTS FROM “TAMING TRANSFORMERS FOR HIGH-RESOLUTION IMAGE SYNTHESIS" ESSER ET AL.

[Late 20 / 2021] Midjourney Beta Blue-Eyed Lucy ≠ Blue-Eyed Lucy

Medium: Midjourney
Focus: Retrying things
Notes: Trying to generate blue-eyed lucy snake (specific morph of ball python) lying in a pile of cigarettes. Keep getting blue-eyed women named Lucy instead. AI completely missing context, fixating on most common usage of individual words. Think the snake + cigarettes = too much semantic bloat for model to handle. (blue eyed + lucy) Model prioritizing human interpretation over literal meaning. Progressively weighting snake.

/IMAGINE A BLUE EYED LUCY LAYING IN A PILE OF CIGARETTES

/IMAGINE A BLUE EYED LUCY BALL PYTHON SNAKE LAYING IN A PILE OF CIGARETTES

[2021] GAS STATION NOCTURNES...AND WHITE SNAKES? HOUR 1

Medium: Midjourney
Focus: Messing around with aesthetic blends, reference guides
Notes: Night playing with "gas station at night, des moines iowa, gaspar noé movie, hyper realistic, cinematic, dirk dzimirsky painting style" prompts. Liminal spaces + night lighting. People generating a lot of horror scenes + dark scenes on blog / midjourney public forum. Do not like that unreal engine 8k image style. Looking for art that feels “real art”. Gaspar Noé + horror aesthetic helped direct model away from weird fixations. Like the idea of exploring suburbia. Trying to add snake back in too soon but unexpected cinematic quality in lighting. Specific contexts >

/A GAS STATION AT NIGHT, IN DES MOINES IOWA, GASPAR NOÉ MOVIE, HYPER REALISTIC, CINEMATIC, DIRK DZIMIRSKY PAINTING STYLE

/A WHITE SNAKE IN A GAS STATION AT NIGHT, CINEMATIC, HORROR STYLE, GASPER NOÉ MOVIE, HYPER REALISTIC

/A SLATEY GREY SNAKE AT A GAS STATION AT NIGHT, HORROR STYLE, HYPER REALISTIC, PHOTO REALISTIC, CINEMATIC, HORROR STYLE

[2021] EVERYTHING IS VAGUELY ASIAN HOUR 2
Medium: Midjourney
Focus: Experimentation fatigue, noticing things
Notes: Noticing all generated text elements are representing Asian-style characters, vaguely Hangul-inspired? Interesting pattern mimicry + dataset training, lots of faux asian gylphs everywhere. If you don’t input model

AN ASIAN MODEL AT A GAS STATION AT NIGHT, SMOKING A CIGARETTE, CINEMATIC QUENTIN TARANTINO MOVIE, HYPER REALISTIC --AR 2:1

AN ASIAN MODEL AT A GAS STATION AT NIGHT, SMOKING A CIGARETTE, CINEMATIC QUENTIN TARANTINO MOVIE, HYPER REALISTIC, SOFT REALISTIC LIGHTING --AR 2:1

AN ASIAN MODEL AT A GAS STATION AT NIGHT, SMOKING A CIGARETTE, CINEMATIC QUENTIN TARANTINO MOVIE, HYPER REALISTIC --AR 2:1

USPCALE RERUN, THIS FELT BIG

THE WORD GAS, GAS

WITHOUT THE WORD “MODEL” IT TENDS TO RENDER OUT A MORE DATED / STYLIZED IMAGE OF AN ASIAN WOMAN WEARING TRADITIONAL OUTFITS

THE CIGARETTE SMOKING COMING IN IN A BIG WAY

[2021] FURNITURE AS WEAPON, WEAPON AS FURNITURE
Medium: Midjourney
Focus: Just messing around
Notes: /Imagine A mac 10 gun designed by hans wegner, with bang and olufsen wood, white background, detailed, 8k. Surprising architectural discipline, materiality + concept together beyond solo style transfer.

[2021] WAIT I’M ASIAN MAYBE THAT’S THE THING TO FOCUS ON HOUR 3
Medium: Midjourney
Focus: Finding focus, subject
Notes: Need a through line to focus the images through abstraction. I think I want to tell the complex story of a midwestern, south korean orphan autobiographically through American realism—an art style I’ve never seen myself represented in. Let’s start here.

LINDEN FREDERICK IS A BIG INSPIRATION HERE, REALISM PAINTINGS REPRESENTING THESE LIMINAL SCENES OF AMERICANA —HOW DO I FORGE A STYLE REPRESENTING ASIAN FEMALE IDENTITY, IN SUBURBIA, WITH THE SAME KIND OF DELIBERATE / INTIMATE / REMOVED TOUCH? LINDEN FREDERICK, NIGHT OWL, OIL ON LINEN

LINDEN FREDERICK, NIGHT STAND [STUDY], 2020
THIS IS MY FAVORITE LIGHTING REFERENCE I’VE BEEN USING

LET’S INTRODUCE FREDERICK’S LIGHTING MORE /LINDEN FREDERICK STYLE

LET’S TRY TO PUT HER IN A CARHARTT SHIRT

AIRPOD MAXES

AIRPOD MAXES, WE GOT THE WORD

SPENT A LOT OF TIME IN A HOODIE LISTENING TO MUSIC OUTSIDE OF A GAS STATION

LOTS OF THINGS GO GOTH QUICK

BETTER

MORE CORRECT

FEELING LIKE A PAINTING 1

FEELING LIKE A PAINTING 2

[2021] THE NEXT DAY, first series
Medium: Midjourney
Focus: End up with a deliberate + semi-cohesive “batch”
Notes: Deliberately leaning into abstraction—matching thematic choices to model's limitations. Model mistakes <> creative opportunities. Fragmented faces + distorted features = fragmented identities. Asian-American experience = inherently navigating between worlds; AI's struggle with coherence = conceptually fitting? Technical weaknesses <> thematic strengths. This is starting to feel like a loosely connected but connected batch of something?

V1, + CHUCK CLOSE FACIAL FEATURES + GRASS

PORTRAIT OF MY BROTHER IN HIS CAR

CAR CRASH

SMOKING CIGARETTE

HOMECOMING

CRYING IN THE HOSPITAL, SAD GIRL STYLE MIXED W/ UNREAL / EPIC + A24 MOVIE POSTER

UPSCALE + PAINTERLY DETAILS IN THE STYLE OF CHUCK CLOSE

90S FASHION MODEL BEAT, OPERATING ROOM BO BARTLETT FACIAL PAINTING FEATURES

UPSCALE, VARIATION + LINDEN FREDERICK LIGHTING APPLICATION

CROSS COUNTRY RUNNER

[2021] AMBIENT MACABRE + MY WAY IS THE WORST WAY
Medium: Midjourney
Focus: Trying the norm-core way of doing things
Notes: It’s obviously better, and I love bats. Redshift + diorama prompt browsing on Reddit. Imagine/ diorama ::2 taxidermy bat carved by transparent glass, glowing inside, atmospheric, 3d render, hyperrealistic, redshift render. The current focus path is definitely the one of most resistance but it’s exciting getting something “real” and less confusing / unsettling to look at (fidelity + style intentionality-wise).

[2022] SECOND series
Medium: Midjourney V 5.1
Focus: Make a series better than the first series
Notes: Less throwing of words into the void and more refining, remixing, and repeating. Prompting is still chaotic but it feels like I know what kind of chaos I’m looking for.

[2022] EVERYTHING FAMILIAR IS SO EASY
Medium: Midjourney
Focus: Playing with the easily recognizable
Notes: Kirby is culturally sticky, placing him in front of the Brooklyn Bridge yields near perfect results after 1 level of upscaling. Familiar IP is frictionless, it’s trained in the model at every level (shape, context, color, emotion). Not generating but recalling, prompt privilege. The shorthand is fun but what about forging more unique styles? Does it mean anything if a generation requires more “work” than another.

READING THIS, MAKING SENSE OF THINGS

NOT EVEN HORRIBLE FOR A FIRST PASS

[2023] THank god for image prompts because midjourney can’t read
Medium: Midjourney V4
Focus: Create and remix surreal environmental scenes
Notes: Remixing visual prompts using vqgan_imagenet_f16_16384 w/ keyword blending. Painting with language—minimal control over textures/worlds but better than nothing. Joined Disco Diffusion Discord. Prompt hacking + collaboration. Prompt engineering = less science, more alchemy. Iteration, randomness, remixing = foundation of AI image generation.

A WHITE ROOM WITH A GLASS TABLE IN THE CENTER
NO IMAGE PROMPTING, BUT ADDING IN A LOT OF DETAILS I’M NOT WANTING HERE

THANKS TO IMAGE PROMPTING, THE STYLE OF TABLE + THE SPACE I WANTED, CLEAN MINIMALIST (BARREN AND CONTAINED SQUARE LIKE MY STUDIO)

THE TABLE IN MY STUDIO

ANOTHER TABLE (LIKE THE ONE THAT USED TO BE IN MY STUDIO), DIFFERENT ANGLES + ALSO WITH GLASS LEGS

[2022] REMOTE CONTROLS + METAL BIRDS
Medium: DALLE 2
Focus: Create and remix surreal environmental scenes
Notes: Training data priorities showing through, styles and consumer products + text well-and better-represented, but complex forms still challenging. Everything struggles so much with mundane objects. Feels like we’re going backwards so going to leave this for now. Text better, but still a lot of bloat and generative hallucinations.

A WOMAN WITH HER FEET UP ON THE COUCH

A COFFEE TABLE WITH DIET COKE CAN, NABISCO OREO THIN WRAPPERS, AND A TV REMOTE SITTING ON IT

GAS STATION COUNTER

VALENTINES DAY GAS STATION PUMP

[2023] “MAKE IT LOOK LIKE SSENSE” + AGENCY SHORTCUTS
Medium: Midjourney
Focus: In achieving editorial qualities for potential professional swipe
Notes: I lead digital for a Soho agency so the stump hands and crazy eyes aren’t giving me a teachable moment. Editorial aesthetics needed. Found adding "SSENSE model" to prompts heightens the editorial style of subject (more Vogue less Better Homes & Gardens). Testing photographer names in prompts. "Tyrone Lebon" = desaturated, pose abstraction. "Petra Collins" = ethereal, Gaussian blur, reliably. Using real photographers' styles—where's the line between reference/appropriation?

[2023] PERFECT IS BORING + DETECTABLE
Medium: Midjourney
Focus: Introducing deliberate flaws for “authenticity”
Notes: The AI perfection is becoming easy to spot, I want to push against the model’s default output and work towards something more imperfect, textured, human, recognizing reality through asymmetries + inconsistencies. Uneven lighting, lens flare, surface flaw, analog texture, “bad” compositions. Not framatic, but residual. Simple + broken context.

[2023] Stable diffusion + control net + character consistency
Medium: Midjourney, Stable Diffusion
Focus: Directional Specifity <> Generational Chaos
Notes: Pose guidance, img2img in Stable Diffusion to refine texture and light. Midjourney for vibe seeding in early stages, SD to turn images from a face to a presence.

[2023] LATENT SPACE AVERAGING + SANDBOX
Medium: Midjourney
Focus: Throwing things into the blender to see what happens
Notes: The tool is still making decisions and you’re just spectating with better taste than it. There’s something nice but also lazy about how /blend doesn’t really even require language.

IMG REF DOG, TRYING TO /IMAGINE MY DOG AS A PUPPY IN A AIREALISM SERIES

DON’T EVEN REMEMBER WHY I WAS MAKING THIS BUT IT WAS A MIRROR FRAME OF WHITE RABBITS

ROAD TRIP, LOFI, WINDOWS 97

CHROME SCULPTURE OF A FEMALE VIDEO GAME CHARACTER HOLDING A GUN

[2024] ASK AND YE SHALL DELIVER
Medium: Midjourney + Stable Diffusion
Focus: Temporary tattoos?
Notes: Last minute ask to create temporary tattoo flash sheet for friend’s birthday party, Libra Ressurection. Midjourney / SD with some quick text lockups and it’s done.

[2025] CARTIER <> SAD BOY ANIME
Medium: Stable Diffusion
Focus: Try and style Cartier imagery into a Neo Yokio-esque anime rip
Notes: The more prompt sensitivity, the more capitalism meets melancholy, the diffused light, soft midtone zone really gives it that anime treatment.

[2025] FOUR HOURS I’LL NEVER GET BACK
Medium: Multi-Platform Flow
Focus: Mini Trailer w/ integrated tools
Notes: In variance complicated but also alarmingly simple workflow using multiple platforms. SD + MJ for base images. Luma/Kling for video. Tried ElevenLabs for voiceover but sounded off— modified my own voice instead (it was giving me a South African accent through11). Animation control lacking, lean toward abstraction. First foray in video. File transfers between tools tedious. Final output on main page.

[2025] THE GREAT GHIBLIFICATION + THINGS ACTUALLY SAY THINGS
Medium: Chat-GPT 4o
Focus: Moore’s Law on Steroids + Quality Leaping
Notes: Industry and timeline hyperventilating over Totoros and considering implications of the future. Micro-asymmetries making everything look less AI. People’s Moore's Law prediction from 2023 completely wrong (mine right). Editorial models now nearly indistinguishable from stock photos, text generation finally works (even across font types).

A PLAIN WHITE TEE SHIRT FLOATING IN THE AIR OVER A WHITE BACKGROUND, EDITORIAL VIBES, PLAIN WHITE BACKGROUND

AN EDITORIAL, SSENSE STYLE PHOTO OF A MAN WITH A BUZZCUT IN A WHITE TEE SHIRT AND ONE CHIPPED TOOTH LOOKING AT THE CAMERA. 90S EDITORIAL STYLE

A GAS STATION SIGN WITH PRICE OF GAS AT 1 DOLLAR

A DISPLAY OF VALENTINE'S DAY FLOWERS AT THE GROCERY STORE CHECKOUT THAT SAYS "HAPPY VALENTINE'S DAY"

[2025] THE RETURN, LIFE + STYLE NO LONGER LOST IN TRANSLATION
Medium: Chat-GPT 4o
Focus: Revisiting American Realism style for my lived experience
Notes: Big difference now: model gets it. Less like prompting, more like art-directing a memory. Not just any Korean-American girl in 1950s scene, but the right one at the right time in the right place. Direct from photographs, or through reference and memory.

[2025] THE BETTY BOOP EXPERIMENT
Medium: KreaAI
Focus: Checking in on 3D capabilities
Notes: Took betty boop screenshot, fed into Krea newly announced 3D model generated, output GLB, converted GLB to USDZ for easy access through Apple’s Quick Look. Lacks sophistication but it’s not soon before long.