Verified cases

Selling the photo shoot away — AI product imagery for cosmetics listings

Open the real page ↗kmong.com · alive as of 2026-08-16

A listing that builds product detail pages for cosmetics sellers. This is the clearest case of what AI actually removed: the studio shoot and the model booking. No revenue figure is published, so we print none — only the platform-displayed transaction count and the published prices.

In plain words

She makes product pages for people selling cosmetics online. Normally that means renting a studio and booking a model. Instead she puts one phone photo into AI and gets magazine-style shots out. The whole cost of the photo shoot disappears. Prices run ₩150,000 to ₩1,350,000.

The whole thing at a glance

  1. To whom

    Cosmetics and beauty sellers…

  2. What

    Content service

  3. Found how

    Kmong — marketplace search p…

  4. Paid how

    Three fixed packages

Flow: To whom, then What, then Found how, then Paid how

The money path, in real screenshots

Screenshots we took ourselves. Each carries the date and the address it came from — don't take our word for it, open the page.

  1. Step 1. What it costs

    ↔ Swipe sideways on a narrow screen

    Same three tiers, a much steeper top. ₩150,000 / ₩450,000 / ₩1,350,000. The bottom tier is “two AI images.” The top is “conversion-led planning plus ten sections.” What raised the price is the planning, not the AI — the moment the deliverable shifts from “make it” to “make it sell,” the number goes up ninefold.

    Captured 2026-08-16 · it may have changed since

  2. Step 2. Did anyone buy

    ↔ Swipe sideways on a narrow screen

    23 orders, 99% satisfaction. And here is where it diverges from the previous case — this is a business account, so tax invoices are available. Business customers ask for them. If that box says “not available,” those orders never arrive in the first place. Registering as a business is a precondition, not a later step.

    Captured 2026-08-16 · it may have changed since

These screenshots are quoted for analysis; the underlying pages belong to their owners. We left out faces and the sellers' education and career panels — what we are reading is the price and the order record, not the person.

Founder
Youngji Yoon (Studio Bob)
Country
KR
Team
—
Startup cost
—
To first revenue
—
Automation
mid
Reported revenue
Revenue not disclosed

What is sold

Three fixed packages (images only / images plus page / full package)

STANDARD ₩150,000 (2 AI scene shots) / DELUXE ₩450,000 (4-section page) / PREMIUM ₩1,350,000 (10-section full package with GIFs). Checked 2026-08-15.

Who buys

Cosmetics and beauty sellers on Korean marketplaces, and small brands

What arrives is not what you keep

  1. What the customer paidThe number on the invoice
  2. AI toolsA share of the monthly subscription
  3. Platform feeDiffers by platform
  4. TaxDiffers by country and business form
  5. What is leftAt or below zero, it is not a price
The rows are not scaled to amounts. Tools, fees, and tax differ per person, so we do not draw a share we did not measure — put your own numbers in the calculator.

This case does not publish its own tool costs, fees or tax. So we show the order and the items only — put your own numbers into the calculator.

AI's part and the person's part

Already had

Eight-plus years in design — agency art director, startup design lead — per the profile. Twelve portfolio pieces are published with client names.

If this is large, the AI is not the reason

What the AI did

Takes a single product photo as input and generates AI scene and model shots from it, replacing the studio shoot and model booking. The input being one phone photo is the point — the buyer has almost nothing to prepare.

the part a tool replaced

What the person did

Reading the brand concept, planning the page, retouching and compositing the AI output, and laying out sections. The listing itself says an art director retouches the output rather than shipping raw generations.

stories that omit this do not replicate

The three columns are not proportional to contribution — we did not measure that. This is a schematic for separating them.

How customers were found

  1. Kmong — marketplace search placement

  2. Twelve public portfolio pieces, client names attached

  3. Platform-displayed “replies within 1 hour”

  4. Paying customers

    STANDARD ₩150

Channels we could confirm from public sources. Conversion rates were not disclosed, so they are not shown.

Flow: Kmong — marketplace search placement, then Twelve public portfolio pieces, client names attached, then Platform-displayed “replies within 1 hour”, then Paying customers

How we verified this

Evidence strengthB· 1 source(s)

Public dashboard or platform figure. May be self-reported

Checked the Kmong listing directly on 2026-08-15. Platform-displayed values — 23 completed transactions, 99% satisfaction, 13 reviews averaging 4.9, corporate account. The seller's real name is public on the profile. No revenue figure is published, so we print none. ⚠️ One review ends by recommending a *different* marketplace, which suggests it was pasted from elsewhere — a good illustration of why we do not treat review text as evidence. The listing does not name the AI tools used, so that field is left empty.

See the grading rules

What made it work

What AI replaced is not craft but a fixed cost. Once the studio, model and lighting drop out, the buyer stops comparing this against other designers and starts comparing it against a shoot quote. Changing the comparison is what makes the price work. The art direction is still done by a person.

Risks

  • Existing design skill carries much of this. Generating an image is easy; getting to “a brand can ship this today” is what eight years of experience buys.
  • A four-star review notes the product packaging was hard to render — results vary by product shape.
  • The low barrier cuts both ways: similar listings multiply quickly and compete on price.

Does this work in Korea?

The buyers are domestic sellers, so it applies directly. Note that cosmetics pages fall under Korean cosmetics advertising rules — efficacy wording in the imagery is ultimately the brand's liability. Whether AI-generated imagery must be labelled depends on the platform and the brand, so it is safer to settle that before the contract.

Replicability in Korea4/5
Difficulty3/5

Solo-feasible: partially · Automation: mid

7-day action plan

  1. Step 1. Shoot one product on your phone, generate scene shots from it, and count the attempts until one is usable

  2. Step 2. Multiply that count by your tool costs to get a per-image cost

  3. Step 3. Write down the prices of ten comparable listings and compare against your cost

30-day action plan

  1. Step 1. Narrow to one product type and build three portfolio pieces yourself

    with no client

  2. Step 2. List the service with those three and sell only the smallest package first

  3. Step 3. Count the revision rounds on the first order, and put that number into the next price list

Run the numbers

We do not print a single projected figure here. Put your own numbers in instead.

Break-even calculator →

How this reads for you

Answer 13 questions and we work out what this case would take in your situation. Answers stay in this browser.

Find models that fit you

Get the next verification result

We check the AI income stories people are talking about, grade the evidence, and send you what we found — including what we could not confirm.

Income models this case supports

Related cases

Words you may not know

Verification grade
How far we checked this case. A means the revenue was confirmed by someone other than the founder; B means the business exists but the revenue is only the founder's word.
Revenue
Everything that came in. Tools, fees and tax still have to come out before it is what you keep. The two are not the same.
MRR
Money that arrives every month, as opposed to a one-time payment.
Self-reported
A number the person stated about themselves. Nobody else confirmed it, which is why we do not print those numbers.
Replicability
How likely someone else is to get a similar result. The more the founder already knew the field, the harder it is to copy.