Vetra AI · Case Study 02
Kate Xu · Vol. 02
AI Studio · vetraai.com
“Am I in slides
or video right now?”
The most-asked question in user testing — and the redesign that made the product answer it before anyone asks. Product + landing, shipped in two weeks.
Kate Xu·Product Designer · Aesthetic System·2026
A case study on orientation
— 2026 —
Context · What Vetra Is
02 / 16
Screenshots in. Stories out.
An AI studio for
founders’ storytelling.
The product
Turns founders’ screenshots into launch-ready slides and demo videos.
Platform
The audience
Startup owners who want storytelling that impresses — without becoming editors.
Users
My scope
Product redesign, landing redesign, and the aesthetic algorithm behind every generated visual.
Role
Shipped in two weeks · product + landing + aesthetic system
Context
Research · Signals
03 / 16
Six patterns, every session
What the research kept saying.
01 · Loved
The two-way convert
Start a video, turn it into slides mid-flight — or the other way around.
02 · Video first
Founders start in video
Then want to explore further from what they built.
03 · Undiscovered
Convert went unfound
Most users never discovered the feature at all.
04 · Scattered
Controls everywhere
No mental model of where to control what.
05 · Ambiguous
One selector, two meanings
Pages in slides, story covers in video.
06 · The question
“Am I in slides or video?”
Asked in every single session.
Excitement in every session · confusion in every transcript
Research
Research · The Read
04 / 16
How to read a test session
“Test users perform politeness.”
People smiled at every result — and kept asking where they were. The questions they ask are better data than the excitement they show.
Smiles are noise·Questions are signal
Page 04 · The methodological turn
— · —
Diagnosis · Root Cause
05 / 16
Why the confusion existed
One template, two readings.
From the inside · the founder’s case
A reasonable engineering trade
- One universal layout across slides and video
- Instant familiarity between formats
- A simpler back end
- Easy copy-paste from format to format
From the outside · the user’s experience
Two identical rooms, no signage
- A vertical selector that means pages in one room
- — and story covers in the other
- The only orientation signal, erased
- “Where am I?” in every transcript
A convenience inside is an ambiguity outside
Diagnosis
Evidence · Before
06 / 16
The original design, recorded
Two rooms, no signage.
Same layout, same selector · only the meaning changes
Evidence 01
The Principle
07 / 16
One sentence drove every decision
The layout tells you where you are.
If a screen needs a label to explain which product you’re in, the screen is the problem. Each room’s architecture became the signage.
The founder signed off · v1 shipped
Principle
Solution · The Redesign
08 / 16
Three moves, one mental model
Rooms. People. Doors.
01
Rooms
Each format gets its own room. Slides keeps the deck-builder model everyone owns — a vertical page selector. Video gets parallel story rows: one row per storyline, never an editor timeline.
02
Modes
Modes categorize the person, not the format: Yolo, Director, and Showcase run on one universal parameter template — switching intent never costs the familiarity of the controls.
03
The Door
Convert became a visible button — a door between rooms instead of a hidden passage. Start in video, walk into slides mid-flight. The favorite feature, finally findable.
The screen answers “where am I” before anyone asks
Solution
Solution · The Logic
09 / 16
Old model → new model
The same product, remapped.
Structure
One universal layout shared by both formats → Two rooms, two layouts — the room is the signage
Old → New
“Mode”
The format — a Slides ⇄ Video switch → The person — Yolo · Director · Showcase
Old → New
Selector
One selector, two meanings — pages here, story covers there → One meaning each — pages in Slides, story rows in Video
Old → New
Convert
A hidden passage, buried among parameters → A visible door between rooms, both directions
Old → New
Nothing was removed. Everything was re-addressed.
Page 09 · The remapping
Old vs New
Evidence · Convert
10 / 16
The favorite feature, finally findable
From hidden passage to door.
Start a video · walk into slides mid-flight · or back
Evidence 02
Evidence · Shipped
11 / 16
V1, in motion
Inside the new rooms.
One row per storyline · never an editor timeline
Evidence 03
Solution · The Surface
12 / 16
One design system, two screens
The promise and the product
look like the same thing.
The web workstation and the mobile composer share one design system — so the first session doesn’t break the spell the landing cast.
Two rooms · three modes · one door.
— The whole mental model, in one line
Page 12 · As shipped
— · —
Solution · The Flow
13 / 16
A founder’s path
One flow, not three projects.
From screenshots to story
01
Upload
Screenshots go in — the raw material.
02
Pick a room
Slides or Video — the layout confirms the choice.
03
Choose a mode
Yolo, Director, or Showcase — who you are today.
04
Build across rows
Elements copy-paste across storylines.
05
Convert · export
Walk through the door, or ship it.
The AI stays a polish pass inside each room — not the primary control surface.
Structure for creating · AI for finishing
The Flow
Beyond the Surface
14 / 16
The aesthetic algorithm
Teaching machines taste.
Every visual Vetra generates runs through rules I wrote — human aesthetic judgment, decoded into decisions a machine can execute.
Mood
What a story’s atmosphere implies for its palette, type, and pacing.
Craft floors
The rules that keep generated output from feeling generated.
Proprietary
The rulebook stays client IP — the field is where my practice is heading.
Every generated visual runs through rules I wrote
The System
What’s Next
15 / 16
V1 is a hypothesis
Ship small. Let the data steer.
Signal 01
Where-am-I questions
Do they disappear from session transcripts now that the rooms look different?
Signal 02
Convert discovery
How many users find the door unprompted — and how early in their first session?
Signal 03
Round trips
Do people actually work across both rooms once they can see the path between them?
We deliberately changed no more than the mental model demanded. The next iteration follows the data — not individual conviction.
Three signals decide the next iteration
Next
Closing · Reflection
16 / 16
Reflection
The screen is the signage.
When users ask where they are, don’t answer with a label. Answer with architecture.
01
Layouts orient, labels don’t — identical rooms erase the user’s map.
02
Modes describe people — the format is a room, not a mode.
03
Politeness hides confusion — weight the questions over the excitement.
Kate Xu · Product Designer·katexu.com
End of case study 02 · Vetra AI
— Fin —