creative
Real clothes, AI models: making a Kowloon fashion film
We dressed fictional Kowloon residents in Dries Van Noten and gave them ordinary things to do. The agents produced 24 five-second video candidates; Egor cut selected moments with archival Hong Kong footage into a 47-second film. The arcade model gets less than a second in the final edit.
The video-generation estimate came to US$23.76. Agents handled reference files, scene versions, generation jobs and downloads. We directed the people and gestures, reviewed the results and finished the film in Final Cut Pro. Below, you can compare the real clothes with what the models made of them.
An independent Lyfar Studio fashion study. The Dries Van Noten title and referenced clothes are part of the creative concept; this is not a commissioned or affiliated brand campaign. The figures below cover video generation, with the full exclusions beside the cost table.
Why Dries
We chose the Dries Van Noten AW26 collection because we think it’s a masterpiece. The colours, the layering, the way several patterns work in one outfit. We wanted to see those clothes in this film.
The technical justification is good, too. This collection gives an agent plenty to keep track of: embroidery, checks, floral prints, marbled velvet, different textures and awkward little details. The green checks need to stay green checks when someone turns a shoulder. The orange ribbing needs to stay attached to the right jacket. A convincing face won’t rescue a coat the model has redesigned.
That made Dries a useful stress test for the whole workflow: finding the right references, combining them into a look, generating a first frame, then trying to keep the clothes consistent through movement. We wanted to see where the detail survived and where it wandered off.
The first version of our Fidget Book experiment also used Dries references (choose V1 in the demo). We keep coming back to Dries. They’re cool.
Give the models something to do
We wanted contemporary fashion inside a historical Kowloon Walled City setting: strange haircuts, expensive shoes, people getting on with their day. One carries a stack of red stools. Another weighs dried peel behind a shop counter. Two watch a small television.
Carrying three stools gives you weight, a grip and a reason to move through a room. It also gives the video model several opportunities to get the hands wrong. That’s something you can inspect and direct.
We set the art department around 1989 and mixed in 2026 clothes, plus early digital cameras and keypad phones. The collision was deliberate. The invented shops and characters aren’t a historical reconstruction.
The cast grew to 12 fictional Asian adults, all specified as 19. We asked for freckles, uneven features, visible skin texture, mildly prominent ears and haircuts with character. Red, platinum, blue, cropped green. Each identity had a description and reference frame, so changing a coat wouldn’t mean casting a new face.
Start with clothes you can point to
We downloaded official product photographs and recorded their product links, then mixed pieces into 19 looks, including Adidas shoes and Miu Miu mules. A product photograph lets you specify the collar, pockets, hem and colour relationships. “Expensive fashion” leaves those decisions to the image model.
For the stool scene, we combined a plum embroidered bomber , wool skirt trousers and Miu Miu leather mules .
In the herbal shop, we layered a red marbled velvet dress over a teal printed silk blouse . For the arcade, we combined a checked seersucker bomber , a floral wrap skirt and low ballerina sneakers.
You can recognise the styling in the generated frames, but the model redraws the garments. Fine prints drift. A silver trainer can come back off-white. We keep the product images beside the results so you can see those changes for yourself.
Make the first frame before you buy movement
We used the GPT image-generation tool available inside the agent session for the stills. The tool didn’t expose a model version, so we can’t give you an honest version number. We saved 34 frame versions across 30 scenes.
Each image in a reference pack had a job: a face for identity, product photographs for clothes, an architectural reference for surfaces and room density. We told the model which job belonged to which image. That included spelling out that the catalogue model’s face wasn’t part of the wardrobe reference.
The image tool accepted at most five references in this session. We chose that short ingredient list before asking the model to combine it.
We asked for clean 4:3 frames with room around the hands and feet. You need to see whether a hand holds the scoop before you ask it to pour. Tape noise and camera shake make that inspection harder, so we left the analogue treatment for the edit.
The first-frame brief describes the person, clothes, starting pose and camera position. A roof-walking shot needs somewhere to walk; leave room for the next action before generating the image.
Give the agent a scene record
Codex coordinated the research, file handling and generation calls. We used scene IDs, cast IDs, wardrobe records and a source register. The project ended up with 33 scene records and 56 registered references. Some scenes stayed at the brief or still stage.
The herbal shop is S31. Its first image is v001; its first motion take is m001. Each take points back to the exact frame it used. We saved the prompt, input references, model identifier when available, parameters and output receipt together.
- 01Find garment and place references
- 02Write the scene and cast brief
- 03Generate a clean first frame
- 04Inspect the face, clothes and grip
- 05Save the frame version
- 06Submit and track the motion job
- 07Download, check and catalogue
- 08Select and edit in Final Cut
Revisions got new filenames. We kept the earlier image and its original brief, so we could compare a new look with the previous version without reconstructing a conversation.
The agents could research clothing while Egor reviewed scenes, and prepare more stills while video jobs ran. A private gallery brought the references, frames and clips together for review, including from a phone. It read the scene records, which saved us maintaining a second handwritten shot list.
For video, the agent uploaded the first frame, submitted the job and saved its prediction ID. It checked that same ID until the job finished, then downloaded the output. After an uncertain response, the next step was to reconcile the job state before creating another job. Otherwise, you risk paying twice for the same request.
We kept failed attempts in the records to explain missing clips and avoid repeating work.
Direct one action across five seconds
We used Replicate for the motion jobs. Wan 3 produced 22 of the 24 completed clips. We also completed one Runway Gen-4.5 test and one Seedance 2.5 test.
The still already contained the face, clothes and room. The motion prompt concentrated on the change over five seconds, with instructions to preserve those details and hold one continuous shot.
For the herbal shop, the saved direction included this passage:
Starting from the exact scoop position in the image, the woman moves the scoop a short distance to camera-left until it is above the ingredient pan. She gently tips a few small pieces of dried peel onto the existing pile, then holds the empty scoop above that pan.
“Starting from the exact scoop position” mattered. The generated still had placed the scoop higher than the original brief described. We directed from the image we had.
We kept the actions small: two steps with the stools, a glance at a screen, a short pour. Our first motion pilot showed hanging clothes. We wanted people, and that pilot told us nothing about faces or hand movement. We changed the test. Start with the gesture your film depends on before generating a large batch.
The references and generated scenes
The gallery includes all 30 garment and shoe references and all 34 generated frame versions across 30 scenes. The source photographs sit beside the generated images, with links back to the original product pages. We kept the revisions so you can compare the attempts as well as the finished-looking ones. The generated clothes are interpretations; the photographs show what we asked for.
Each row shows the references for the styled look, followed by its generated frame and motion take where one exists. Some clothes sit outside the crop. We also reused earlier generated frames for identity and wardrobe continuity; these strips are not a list of images attached to every generation call.
S01 · Laundry above the city
01 · Real garment references
Product photographs: Dries Van Noten, Adidas. Individual photographers not listed on the source pages. Garment names link to the original product pages.
S02 · Last bundle
01 · Real garment references
Product photographs: Dries Van Noten, Adidas. Individual photographers not listed on the source pages. Garment names link to the original product pages.
S03 · One tile
01 · Real garment references
Product photographs: Dries Van Noten. Individual photographers not listed on the source pages. Garment names link to the original product pages.
S04 · Fringe between customers
01 · Real garment references





Product photographs: Dries Van Noten, Adidas. Individual photographers not listed on the source pages. Garment names link to the original product pages.
S07 · Clothes without their owners
01 · Real garment references






Product photographs: Dries Van Noten, Adidas. Individual photographers not listed on the source pages. Garment names link to the original product pages.
S08 · Two faces in the mirror
01 · Real garment references




Product photographs: Dries Van Noten, Adidas. Individual photographers not listed on the source pages. Garment names link to the original product pages.
S09 · Let it dry
01 · Real garment references




Product photographs: Dries Van Noten. Individual photographers not listed on the source pages. Garment names link to the original product pages.
S10 · The next stack
01 · Real garment references
Product photographs: Dries Van Noten, Adidas. Individual photographers not listed on the source pages. Garment names link to the original product pages.
S11 · Not asleep yet
01 · Real garment references
Product photographs: Dries Van Noten. Individual photographers not listed on the source pages. Garment names link to the original product pages.
S12 · Left by the door
01 · Real garment references






Product photographs: Dries Van Noten, Adidas. Individual photographers not listed on the source pages. Garment names link to the original product pages.
S13 · Late through the passage
01 · Real garment references




Product photographs: Dries Van Noten, Adidas. Individual photographers not listed on the source pages. Garment names link to the original product pages.
S14 · Yan above the rooms
01 · Real garment references
Product photographs: Dries Van Noten, Adidas. Individual photographers not listed on the source pages. Garment names link to the original product pages.
S15 · Three across the roofs
01 · Real garment references






Product photographs: Dries Van Noten, Adidas. Individual photographers not listed on the source pages. Garment names link to the original product pages.
S16 · Between customers
01 · Real garment references
Product photographs: Dries Van Noten. Individual photographers not listed on the source pages. Garment names link to the original product pages.
S18 · You coming?
01 · Real garment references




Product photographs: Dries Van Noten, Adidas. Individual photographers not listed on the source pages. Garment names link to the original product pages.
S19 · Two bowls, no conversation
01 · Real garment references




Product photographs: Dries Van Noten, Adidas. Individual photographers not listed on the source pages. Garment names link to the original product pages.
S20 · Listening through the wall
01 · Real garment references
Product photographs: Dries Van Noten, Adidas. Individual photographers not listed on the source pages. Garment names link to the original product pages.
S21 · Strong tea, no hurry
01 · Real garment references
Product photographs: Dries Van Noten, Adidas. Individual photographers not listed on the source pages. Garment names link to the original product pages.
S22 · One more fishball
01 · Real garment references



Product photographs: Miu Miu, Dries Van Noten. Individual photographers not listed on the source pages. Garment names link to the original product pages.
S23 · Five centimetres of cloth
01 · Real garment references



Product photographs: Dries Van Noten, Adidas. Individual photographers not listed on the source pages. Garment names link to the original product pages.
S24 · Hold the line
01 · Real garment references
Product photographs: Dries Van Noten. Individual photographers not listed on the source pages. Garment names link to the original product pages.
S25 · Through the lightwell
01 · Real garment references
Product photographs: Dries Van Noten, Adidas. Individual photographers not listed on the source pages. Garment names link to the original product pages.
S26 · Last shop open
01 · Real garment references




Product photographs: Dries Van Noten, Adidas. Individual photographers not listed on the source pages. Garment names link to the original product pages.
S27 · Three stools home
01 · Real garment references



Product photographs: Dries Van Noten and Miu Miu. Individual photographers not listed on the source pages. Click a garment name for the original product page.
02 · Generated first frame · S27

03 · Five-second motion take
Read the saved motion direction
The woman takes two short grounded steps toward camera-right along the passage, carrying the compact nested stack of three red stools at the same central waist position shown in the image. Both hands maintain their original grips.
Excerpt from the submitted prompt. Wan 3, requested 1080p, five seconds. First frame v001; motion take m001.
S28 · Air from the next room
01 · Real garment references



Product photographs: Dries Van Noten. Individual photographers not listed on the source pages. Garment names link to the original product pages.
S29 · The small screen
01 · Real garment references






Product photographs: Dries Van Noten, Adidas, Miu Miu. Individual photographers not listed on the source pages. Garment names link to the original product pages.
S30 · One more credit
01 · Real garment references



Product photographs: Dries Van Noten. Individual photographers not listed on the source pages. Click a garment name for the original product page.
02 · Generated first frame · S30

03 · Five-second motion take
Read the saved motion direction
The woman slowly turns her eyes and head from the doorway back toward the arcade screen. Her joystick hand nudges the red ball joystick once, while her other hand taps one button without changing its wrist position.
Excerpt from the submitted prompt. Wan 3, requested 1080p, five seconds. First frame v001; motion take m001.
S31 · Paper packet
01 · Real garment references



Product photographs: Dries Van Noten. Individual photographers not listed on the source pages. Click a garment name for the original product page.
02 · Generated first frame · S31

03 · Five-second motion take
Read the saved motion direction
Starting from the exact scoop position in the image, the woman moves the scoop a short distance to camera-left until it is above the ingredient pan. She gently tips a few small pieces of dried peel onto the existing pile, then holds the empty scoop above that pan.
Excerpt from the submitted prompt. Wan 3, requested 1080p, five seconds. First frame v001; motion take m001.
S32 · Before the lantern closes
01 · Real garment references



Product photographs: Dries Van Noten, Adidas. Individual photographers not listed on the source pages. Garment names link to the original product pages.
S33 · Your serve
01 · Real garment references





Product photographs: Dries Van Noten, Adidas. Individual photographers not listed on the source pages. Garment names link to the original product pages.
02 · Generated first frame · S33 · v001

03 · Five-second motion take · m001
Notes on the catalogue references
We built the wardrobe from the AW26 collection pages and the linked product photographs. The boxy red sweater and striped polo appeared in that collection, while their product descriptions named Spring–Summer 2026. The embroidered white shirt’s title and description also disagreed on the fabric. We used the visible garments and kept those metadata conflicts in the source records. Some shirts, ties and skirts follow styling visible in a catalogue photograph without a separately verified product page.
What the video generation cost
These are published-rate estimates checked on 22 October 2026, calculated from the saved project jobs. We haven’t reconciled them against an invoice.
| Video model | Completed clips | Settings and listed rate | Estimate per 5-second clip | Subtotal |
|---|---|---|---|---|
| Wan 3 | 22 | 1080p, US$0.20/second | US$1.00 | US$22.00 |
| Runway Gen-4.5 | 1 | US$0.12/second | US$0.60 | US$0.60 |
| Seedance 2.5 | 1 | 720p, non-video input, US$0.2312/second | US$1.156 | US$1.156 |
| Total | 24 | 120 seconds of requested candidate footage | US$23.756, rounded to US$23.76 |
Our Wan setup cost an estimated US$1 per five-second candidate. A usable shot can cost more if you need several attempts.
Six failed video predictions also appear in the ledger. We assigned those zero in the estimate under Replicate’s failed-run billing policy . We counted unique prediction IDs, so an original MP4 and its silent editing copy didn’t become two paid generations.
The image tool exposed no per-image charge or subscription allocation. Image generation, agent and tool subscriptions, our time and editing are outside this figure. Taxes, discounts and credits aren’t reconciled either. US$23.76 covers the estimated video generation. Making the whole film involved the rest of that work, too.
Inspect the motion
We checked the downloaded files with ffprobe, decoded them with ffmpeg and reviewed six sampled frames per clip. Those checks caught technical failures and visible changes; they don’t prove that every moment moves well. You still need playback and an edit decision.
The Runway manicure test inserted a cut around 2.79 seconds into a shot that should have held. In the lantern study, the model lowered a paper strip onto the bench instead of attaching it to the lantern. Egor kept a short part of that take in the film, using the gesture the model produced. The full clip still fails the original direction.
We preserved the provider originals and made silent editing copies without re-encoding the video.
Cut the moments in Final Cut Pro
The saved Final Cut project references 22 of the 24 generated takes, alongside three archival footage sources. Egor cut in a 1080p, 24fps project and exported the 46.625-second film at 720p.
These three wardrobe examples each started as a five-second generation:
| Scene | Starts in the film | Length used |
|---|---|---|
| Carrying the red stools | 11.83 seconds | 1.29 seconds |
| Arcade | 19.38 seconds | 0.92 seconds |
| Herbal shop | 20.29 seconds | 1.17 seconds |
The arcade model’s face, clothes and hand on the controls needed to read for 22 frames. The longer source take gave us room to choose that moment in the timeline.
Egor kept the finishing in Final Cut. The project has a Dehancer Pro v7 adjustment layer across the film, Pro Zoom on the money-counter shot, and shake and glitch effects on selected passages. Keeping those treatments in the timeline let him tune them against the archive footage after choosing the cuts.
The archive inserts, music and street ambience belong to this editing stage. The automated generation supplied candidate footage; Egor chose how it sat beside the existing Hong Kong images and where to cut.
Footage and sound
These are the source files referenced by the saved project:
- Archive: Life in Kowloon Walled City , Kowloon Walled City: “And the Walls Came Tumbling Down” (1984) , and Chika Asamoto in Hong Kong, 1988: STRAY .
- Music: “Pump That Shit Up” by Jammin The House Gerald .
- Street ambience: Walking in Hong Kong Causeway Bay: Immersive Street Sounds Walk , matching the named ambience file in the project.
Start with one garment photograph, one character and one action. Save the sources and brief together. Generate a still, inspect the face and grip, then freeze that version and animate five seconds. Watch it in a timeline before commissioning the next batch.
Bring that first scene to an AI workshop . We can build the reference and generation workflow around a shot you want to use.
Interested in this?
Learn about Creative Production →






























