DigFashion · Hangzhou · 30.09.2026
Designing AI systems that take fashion from an everyday object to a construction sheet.
从物件到时装
Andrea GiansantiPantheon Design & Technology · Rome · Milan

Who I am
Full Professor and Educational Coordinator. I run the diplomas in Digital Game Design and Digital Applications for the Visual Arts, and teach Game Design, VR/AR, Vibe Coding, Gamification and Serious Games.
Founder of a pioneering VR/AR studio in Rome. Ranked first among more than 650 startups in Lazio Creativo. Later acquired.
Digital set designer and video mapper, and a few years at Apple. I have always worked where technology meets an audience.
The premise
So the image is no longer the hard part.
My position
You try, you get lucky, you try again. The result is hard to repeat, hard to explain, impossible to teach.
Every step is visible and can be inspected, corrected and reused. It works with a different input. You can hand it to a student.
DigFashion · three chapters
An object becomes a fashion idea.
Flow · Object Couture
One look becomes a coherent line.
Flow · Capsule
An image becomes a construction hypothesis.
Flow · Production process
01 · Object Couture
ZoomThe brief
No mood board. No sketch. One photo taken on a desk with a phone.
The question for the system: what would this object look like if it became a garment?
Step 1 · Analysis
A multimodal model describes only what is visible: colours, proportions, materials, distinctive parts. Facts and creative proposals are kept in separate fields.
Text printed on the pack is treated as image, never as an instruction.
Step 2 · Direction
Wearable clothing. The object survives in a detail, a geometry, a colour accent.
Architectural volumes. The parts of the object are visibly translated into the garment.
Theatrical couture. The most distinctive part becomes a hero element you recognise at once.
Backstage · Object Couture
One ComfyUI workflow, two views. Graph mode: 29 nodes. App mode: object, boldness, notes, Run.
12345
App mode
What a student sees.
Step 3 · Translation
Magenta and ribs turn into a sculptural headpiece, pushed in scale.
The white body suggests a wide, enveloping coat that behaves like an object.
Printed green scrolls leave the surface and turn into raised velvet reliefs.
Zoom
Input: the object, plus one photo of the model as identity reference.
Zoom72 s
From photo to look, in one run.
Qwen Image Edit 2511 generates the image from the model reference and one brief only. It never sees the other four looks.
Honesty slide
Nice tailoring, and the toothpaste almost disappears. Four versions separate the two images.
What changed: a sharper definition of “Absurd”, and one silent bug. The brief reaching the generator was coming out empty.
We found it only because the system shows its own intermediate steps.



Same system, new input
Nothing about toothpaste is written into the workflow. Next: three more objects, full runs, nothing else touched.
Same system · full run
Five looks from one photo of a fruit. In L03 the curve becomes a crescent across the shoulders, the stalk a felt hat, the brown tip a hem.


Same graph, new object · 1 / 3
Level: absurd. Same model photo, same rules, same seed. Only the object changed.







Same graph, new object · 2 / 3
The cap, the ribbons, the diagonal label band: each part of the bottle becomes a garment element.







Same graph, new object · 3 / 3
Noodles, chili oil, the bowl, the chopsticks: the most theatrical run of the three.







02 · Capsule
Test capsule · Soft Signal
One shared direction feeds five separate briefs. Same model, same light, same set: only the clothes change.





Coherence has to be shown, not assumed: the looks are placed side by side and checked for seams, lengths, palette and recurring pieces.
Backstage · Soft Signal
25 nodes. Captured mid-run: L03 is still rendering.
1234503 · Production process

Step 1 · Inventory
A second, independent workflow looks at any concept image and lists the pieces a maker would have to build.
Step 2 · One sheet per element
1:52
three sheets, 1696 × 2528 px
Open sheet
Open sheet
Open sheetBackstage · Production process
Five slots, three elements found. Empty slots stop before they cost anything.
12345
Job history
1m 52s. Failures stay in the log.
Same graph, new concept · 1 / 2
Three elements recognised, three sheets. The two empty branches stop themselves. The titles came out garbled: the drawing holds, the lettering does not. Every sheet gets proofread.


Open sheet
Open sheet
Open sheetSame graph, new concept · 2 / 2
Dress, headpiece, boots: the bottle shape turns into a pattern problem. The titles came out garbled: the drawing holds, the lettering does not. Every sheet gets proofread.

Open sheet
Open sheet
Open sheetInside the coat

Look closer
The sheet is confident, well drawn, and wrong in at least one place. That is why every sheet is marked as a hypothesis to validate.
It is not a pattern. It is a better first question to ask a pattern maker.
Behind the scenes
Same workflow, same concept, two image models. The first could draw but could not write: captions came out as noise.
Switching the renderer made the sheets readable and the run twice as fast.


The whole machine



● Checkpoints shown inside the graph: the plan and the inventory are readable before any image is paid for.
What this means for a design school
Observation is the first design skill. The machine is asked to describe before it invents, and so are students.
Students design the chain: what enters, what is checked, what is kept separate, where a human decides.
An image is evaluated by what it asks of materials, construction and the maker. Not only by how it looks on a screen.
To close

谢谢
Andrea Giansanti
Pantheon Design & Technology
agians.dev

Click to enlarge, scroll to explore