If you want a GRWM outfit image where the clothes are the point, name every garment, the shoes and the bag in the prompt, and expect the model to add, swap or crop a few of them anyway. I opened eight generated images built from that kind of prompt and compared each against the prompt that was supposed to produce it. The big pieces (blazer, bomber, tweed jacket, cardigan, denim skirt) came out as written. The small and structural details did not: shoes changed style, bags were held instead of placed, a "head-to-toe" frame lost the head, and every poster or sign turned into garbled lettering.
This article goes through those eight images one by one. Nothing here is a statistical result: it is eight images, each inspected once, so read the observations as examples of what can happen.
What was checked, and what was not
The eight images are vertical (2176 x 3856 pixels, measured by opening the files), were generated on 2026-07-31 according to the file dates, and carry the label "Krea2 Turbo + SeedVR2 7B" in the original article. The file names contain "seedvr2_7b", which is consistent with an upscaling pass. No generation record with seeds, steps or settings is attached to these files, so the prompts below are quoted as they were published in the original article and I cannot independently confirm that each one produced the image next to it.
What I can confirm is what is visible. I looked at each image for the garments, shoes, bags, hands, and any text. Nothing in the images identifies a real Seoul neighbourhood: the only place names visible are words that the prompt itself asked for, and those are partly garbled (see the mood-board section). When this article says "Seongsu" or "Hongdae", it is repeating the prompt, not something the picture proves.
Mirror selfie: a phone over the face keeps attention on the outfit
Mirror selfie shot type, a Korean GRWM outfit scene in a minimal Seoul studio apartment, full outfit visible from head to shoes in the mirror, oversized oatmeal wool blazer layered over a white ribbed tank, high-waisted straight indigo denim, black leather loafers, small crescent shoulder bag, silver hair clip and thin hoop earrings, relaxed phone-in-hand pose with one knee slightly bent, tidy clothing rack and neutral bedding in the background, soft morning window light, beige cream and denim palette, realistic fabric texture, clean Instagram composition, Sony A7 IV look, 35mm documentary framing
Nearly every item in the prompt is present: the oatmeal blazer with dark buttons, the white ribbed tank, a black crescent shoulder bag with gold hardware, a silver hair clip, a small hoop earring, one bent knee, and both loafers fully in frame. Because the phone covers the face, the viewer reads the outfit first, which is exactly what this kind of post needs.
Two differences are worth noting. The prompt said "straight" denim and the jeans look wide-leg. And the rack to the right holds several jackets plus another pair of black loafers on a shelf, so the image contains shoe-like objects that were not part of the outfit. Small white speckles are also scattered over the wall and rack. The wall poster shows a street photo and a garbled wordmark that looks like "MWG" rather than "GRWM".
Street walk: one named shoe became a different shoe
Street-style full body shot type, Seoul model off duty K fashion outfit during a Hongdae street walk, cropped black bomber jacket over a fitted heather-gray long sleeve top, low-rise charcoal wide-leg trousers with puddled hem, slim black sneakers, worn leather shoulder bag, narrow sunglasses pushed onto hair, silver chain belt detail, subject walking across a painted crosswalk with natural mid-step pose, complete outfit visible head to toe, overcast Korean street light, muted black gray cream palette, documentary street fashion framing, Fujifilm X-T5 color, 40mm lens, crisp garment folds and realistic nylon texture
The silhouette matches: cropped black bomber with visible sleeve pocket, gray top, low-rise wide charcoal trousers, a belt with a silver chain, and a black bag hanging off the shoulder. Both feet are in frame and the mid-step pose has no twisted leg, which was the main risk I would have expected.
Three things did not follow the prompt. The sunglasses are on the face, not "pushed onto hair". The shoes are black suede sneakers with three white stripes, not "slim black sneakers"; the stripes were never requested. The trouser hems reach the shoes but do not obviously "puddle". Being short on detail ("slim black sneakers") left room for the model to fill in a recognisable design of its own.
Cafe look: the skirt stayed off the chair, the shoe did not stay clean
Three-quarter outfit editorial shot type, Korean cafe GRWM outfit in a Seongsu concrete cafe, cream boucle cropped jacket over a pale butter-yellow knit camisole, ivory satin midi skirt, white ankle socks with glossy black Mary Jane shoes, structured mini top-handle bag on the chair, pearl hair pin and small silver watch, seated sideways at a brushed steel cafe table with one hand near an iced latte, outfit visible from shoulder to shoes, calm neutral expression, soft window light, cream ivory butter yellow and black palette, tactile boucle and satin texture, clean Korean fashion magazine composition, Canon EOS R5 look, 50mm natural perspective
This was the strongest garment rendering of the set. The jacket shows a nubby boucle weave with beaded trim and gold buttons, the yellow knit is ribbed, and the satin skirt has believable sheen and folds that do not merge into the wooden chair. A pearl hair pin, pearl earring, silver watch and an iced drink with a straw are all visible, and the hand holding the straw is readable.
The flaws are at the floor and in the details. The left Mary Jane sits against a black fluffy shape that the prompt never mentioned, so the shoe outline is hard to separate from it. The bag is held by its handle beside the chair rather than placed "on the chair". Wall posters contain "GRWM" plus partly readable and partly garbled text, and a strip of glitchy tiles runs along the bottom edge of the frame.
Lookbook: the frame that lost the head
Head-to-toe lookbook shot type, preppy Seoul GRWM outfit on a clean boutique studio set, navy cropped cardigan with gold buttons layered over a white poplin shirt, charcoal pleated mini skirt, sheer black tights, chunky black loafers, burgundy mini satchel, slim ribbon hair tie, outfit centered in vertical frame with arms relaxed and feet slightly apart, complete silhouette visible with negative space around shoes, matte gray backdrop, softbox key light, navy white charcoal burgundy palette, crisp pleats, knit texture and polished leather detail, Nikon Z8 look, 50mm editorial lookbook framing
The garments are clean: pleats read as pleats, the cardigan has a ribbed hem and gold buttons, the white shirt wrinkles naturally, the tights are sheer with visible leg shading, and both loafers are complete. The "slim ribbon hair tie" shows as a cream bow on the hair.
But "head-to-toe" was not honoured. The top of the frame cuts off at the chin, and the arm on the right leaves the frame. The burgundy satchel is held in the hand instead of hanging quietly, and it has a burgundy ribbon tied to the strap, which the prompt did not request. The backdrop is a gray wall with a framed poster that reads "ERV / ARM / GOF" and garbled small text, not a plain matte backdrop. If a real lookbook crop matters, the prompt likely needs explicit headroom wording; I have not tested a fix.
Flat lay: a coherent outfit, plus extras nobody ordered
Product flat lay shot type, Korean GRWM outfit arranged on a pale wood floor before getting ready in Seoul, folded cropped trench jacket in beige cotton twill, white fitted ribbed tee, dark raw denim maxi skirt, black square-toe ankle boots, small olive nylon shoulder bag, silver oval sunglasses, tortoiseshell claw clip, sheer socks, lip tint and phone placed as minor GRWM props, neat grid arrangement with each item separated and fully visible, overhead softbox lighting, beige white indigo black olive palette, sharp textile detail, clean Instagram carousel composition, Phase One product editorial look, 80mm controlled perspective
As a coherence check this is the most useful image, because it shows every item separated. The trench, white tee, dark denim skirt with a side slit, olive bag and tortoiseshell claw clip are all as requested, and there is no pose or hand to go wrong.
Again there are mismatches: the boots have pointed toes and block heels, not "square-toe"; the sunglasses are light gray, not silver; two white books with a "GRWM" cover and garbled captions appear, along with earrings; and the lip product area includes a lip-shaped blob beside a tube. The bag carries a garbled "gg"-like logo. The prompt asked for "each item separated", and that part held, but the set of items expanded past the written list.
Mood-board collage: the style lock that did not lock
Mood-board collage shot type, Korean GRWM outfit planning board for Seoul street style, four small printed outfit references showing a mirror selfie blazer look, Hongdae bomber jacket street walk, Seongsu cafe skirt outfit, and preppy cardigan lookbook, surrounded by fabric swatches of wool boucle satin denim and ribbed knit, black loafers photo, mini shoulder bag clipping, silver hair accessories, handwritten notes for silhouette and palette, Seoul cafe receipt and subway card as context, arranged on a cream desk, soft daylight, neutral black denim burgundy butter-yellow palette, Pinterest-ready but editorial and organized, 24-70mm reference photography feel, crisp paper texture and readable fashion planning layout
The board has what a planning layout needs: boucle, denim and knit swatches, loafers, a black bag, a silver ring and a chalkboard-style card reading "GRWM". Only "Hongdae Look." is cleanly legible; the other handwriting ("Seonsu Buttake", "Andse Wook") and the chalkboard lines are garbled.
The four little photos do not repeat the outfits from the other images: they show a maroon blazer, two maroon bombers and a yellow cardigan over a denim skirt, all in street settings. The prompt's "mirror selfie blazer look" appears as a street phone selfie. The board was arranged on a wall, not a "cream desk". So a collage prompt is not a way to lock a series; it invents its own outfits.
Across all eight: where the prompts held and where they drifted
| Element | Usually held | Drifted in these images | |---|---|---| | Main garments (blazer, bomber, tweed, cardigan, skirt) | Yes, in all eight | Denim cut (straight became wide-leg) | | Shoes | Present and complete in the mirror, street, lookbook and flat-lay images | Style changed (striped sneakers, pointed boots); left Mary Jane merged with a black shape | | Bags | Present in every outfit image | Held instead of placed; extra ribbon | | Framing | Mirror and street complete | Lookbook lost the head | | Text on posters, cards, labels | Words copied from the prompt appeared ("GRWM", "Hongdae Look") | Every other line garbled | | Extra objects | | Additional shoes, books, earrings, ribbon |
Two more images from the same set, a storefront mannequin and an accessories close-up, are not shown but fit the pattern. In the storefront, the hobo bag hangs from the rail instead of the mannequin shoulder, the requested handwritten sale card is absent, and poster text is garbled ("Liierca"). In the accessories frame, the prompt asked for a close-up of objects, yet a woman standing at the table, cropped above her eyes, appears, and an unrequested silver bar sits beside the socks.
What this suggests for your own prompts
What follows is interpretation of these eight images, not a tested recipe.
- Specify the fragile details by type, not category. "Slim black sneakers" produced striped suede sneakers; naming a silhouette (low-profile, plain black, no stripes) is the obvious next step, but it is untested here.
- Put the shoes in the frame description and say what must not appear. In the mirror, street, lookbook and cafe images, where the prompt asked for the outfit down to the shoes, shoes were in frame, but nothing forced the right style.
- Treat all in-image text as decoration. Names like Seongsu or Hongdae in a prompt can leak onto posters as half-formed words and do not make the location recognisable.
- Do not expect a collage to match your other images.
Proposed prompt change (unverified)
This is a proposed edit to the lookbook prompt, not something that was generated:
Head-to-toe lookbook shot type ... full body including the head, headroom above the hair, feet and floor visible, no posters or text in the frame ...Try it in [Prompt Lab](/promptlab/en) and generate in [Create](/create/en). If you then want a larger version of a clean result, inspect garments first and only then use the [Upscaler](/upscaler/en); upscaling sharpens the image that exists, including any odd object beside a shoe.
FAQ
Do Seoul place names in the prompt make the picture look like Seoul?
In these images, no visible detail identifies a neighbourhood. The names showed up only as partly garbled text on posters and notes.
Why do the shoes keep changing?
A short shoe description leaves the model free to choose design details. The three white stripes on the sneakers are an example.
Is one upscale pass worth it?
File names suggest these images went through SeedVR2 7B, but no before-and-after pair is available, so I cannot say what it changed.
> What this article is based on. Eight vertical images (2176 x 3856 px), generated 2026-07-31, labelled Krea2 Turbo + SeedVR2 7B in the original article; all eight were opened and inspected (six shown, two described in text). No generation records were attached, so prompts are quoted from the original article and their match to each image is unconfirmed. Seoul locations are not visually confirmed.