Why I’m building exact product previews instead of AI guesses

The day job

I work as a marketing manager at a web-scraping company. That job taught me SEO the hard way: what people actually search for, what makes a page worth reading, and how fast a nice-looking page stops mattering when it doesn’t answer the question.

Then AI agents got good enough to write real code, and I started building with them. Not demos: things I actually needed. I’m not a trained developer, so I learned the way a lot of us learn now. By asking, by breaking things, and by checking every result myself.

The mug nobody could see

My brother-in-law runs a small family gift shop that prints personalized mugs. Most orders start on Instagram and end on WhatsApp. A customer sends a photo of their mother, a name, a date, and the shop prints it on a mug.

Then the questions start. “Will the photo get cut off at the handle?” “Is the text too small?” “Can I see it first?” The honest answer was usually a quick mockup made in a hurry, or “trust us”. Some customers did. Some didn’t order.

I watched that happen again and again. The product was fine. What was missing was a way to show it before it existed.

So I built a studio

I started with one thing: let the customer see their own mug, from every side, before it’s printed. A 3D mug in the browser, with their photo placed exactly where the press would put it.

Then it grew. Magic mugs, coffee cups, clocks, plaques, pillows, gift boxes, and photoreal images the shop could use in ads without booking a studio every time. I built most of it with AI agents, and along the way I learned where they shine and where they guess.

That studio is now Mockivo.

What AI is great at

To be fair, today’s image and video models are remarkable, and I use them every day. Google’s Pomelli turns a single phone photo into studio and lifestyle shots for free.[3] Amazon’s ad tools generate video ads from product images at no charge to advertisers.[4] Through APIs, an image costs from about $0.007 to $0.21 per try, and an 8-second video from about $0.40 to $4.80 per try, at list prices.[5][6][7]

If you need a good-looking photo of a generic product, use those tools.

Where it guesses

Generative models don’t place your image. They redraw it. That’s what makes them good at mood, and it’s why they’re unreliable for the details a personalized order depends on:

Logos, edges and small text can change from one image to the next.[8]
Sides the model hasn’t seen get invented.[8]
Curved and shiny surfaces are easy to get subtly wrong.[9]
The same prompt gives a different image every time.

For a generic product, a slightly different handle is a harmless variation. For a personalized one, a slightly different face is the wrong product. And the customer usually can’t send it back: EU consumer law and Turkey’s distance-sales rules both exclude goods made to the customer’s specifications from the usual right of withdrawal.[1][2] Mistakes get caught in the preview, or not at all.

A preview is a promise

That’s the idea Mockivo is built on. An exact preview starts from the product, not from a picture of it:

A 3D model of the real product, with its true shape and its true print area, checked by a person.
The customer’s file is placed, not regenerated. Every pixel lands where it will be printed.
Settings, not prompts. The same settings give the same image tomorrow.
A link that carries the design. The shop sees exactly what the customer saw.

That last one changed the WhatsApp conversations more than anything else. When the customer designs the mug and sends the exact result, the chat starts at “yes” instead of “what will it look like?”

AI still has a place: the world around the product. Backgrounds, scenes, mood. Render the product exactly, and let AI build the world around it. There’s a practical reason too. AI providers come and go: OpenAI shut down the Sora API on 24 September 2026.[10] A 3D model of your product stays usable whatever happens to any single model.

Five questions I’d ask any preview tool

Does the customer’s photo come out exactly as uploaded?
Can they see every side, including behind the handle?
Is the print area the one your production really uses?
Will the same design give the same image tomorrow?
Can the customer send you the exact design as a link?

If the honest answer to any of them is “usually”, you’re looking at a guess.

Where I am now

Mockivo is in private beta. I set up each store by hand: AI agents do the heavy lifting, and I check every result. It’s slower than I’d like, and it’s the only way I know to get it right.

If you sell personalized products, try the live demo and tell me what’s missing, or join the waitlist. I read every answer.

Selman

Sources

Links checked on 1 October 2026. Prices are list prices as of October 2026. “Secondary” means a third-party summary.

Directive 2011/83/EU on consumer rights, Article 16(c): no right of withdrawal for “goods made to the consumer’s specifications or clearly personalised”. https://eur-lex.europa.eu/eli/dir/2011/83/oj
Mesafeli Sözleşmeler Yönetmeliği (Turkey’s Regulation on Distance Contracts), Article 15: goods prepared to the consumer’s wishes or personal needs. Official Gazette No. 29188, 27 November 2014. https://www.mevzuat.gov.tr
PetaPixel, “Google’s Pomelli Photoshoot feature is here”, 20 February 2026. https://petapixel.com/2026/02/20/googles-pomelli-photoshoot-feature-is-here-to-hammer-nails-into-the-coffin-of-photography/
Amazon Ads, “Create high-quality AI-generated videos in minutes”. https://advertising.amazon.com/resources/whats-new/create-high-quality-ai-generated-videos-in-minutes
Google, Gemini API pricing (Nano Banana image models; Veo 3.1 video from $0.05/s at 720p to $0.60/s at 4K). https://ai.google.dev/gemini-api/docs/pricing
CostGoat, OpenAI image pricing (gpt-image-2 up to about $0.211 per high-quality 1024² image) (secondary). https://costgoat.com/pricing/openai-images
eesel.ai, Reve 2.1 pricing (about $0.0067 per image) (secondary). https://www.eesel.ai/blog/reve-2-1-pricing
PixReady, “AI product rendering vs professional rendering”. https://www.pixready.com/blog/ai-product-rendering-vs-professional-rendering
Rewarx, “Photoroom vs Pebblely: AI product mockup generation face-off” (secondary). https://www.rewarx.com/blogs/photoroom-vs-pebblely-ai-product-mockup-generation-face-off
NewsBytes, “Sora API shuts down September 24”. https://www.newsbytesapp.com/news/science/sora-api-shuts-down-september-24-what-users-should-do/story