Skip to content
Aziz Belhadj Director · Tunis
AI production2026

Doucet Latendresse: A Catalog That Keeps Growing

Ongoing AI-generated product photography for a Canadian fine jewelry house, delivering a full studio shot set for every new piece without a physical shoot.

AI Case Studies
Produced with AI under human creative direction, hand finished before delivery.

The brief

Doucet Latendresse is a fine jewelry house in Canada, with a catalog running past eight thousand pieces. Getting real, consistent photography in front of every one of them, at the pace new pieces get added, was never going to work as a traditional shoot. This is a long-term, ongoing engagement: a working slice of the catalog today, growing every week, not a fixed batch of images.

The shot list

Every piece gets the same four shots: a front-facing hero, a side profile, a true edge-on profile, and a worn shot on a hand. Same convention whether it’s a ring, a pendant, or a bracelet, so the catalog reads as one coherent collection instead of a patchwork of styles.

Fidelity, made visible

Every piece below shows the client’s own reference photo next to the AI-generated hero shot built from it, followed by the other three angles the pipeline produced. What starts as one catalog photo becomes a full studio set, matched stone for stone, prong for prong.

Rings

130AP09611
130AP09611 -- Client photo
Client photo
130AP09611 -- AI-generated
AI-generated
130AR70110
130AR70110 -- Client photo
Client photo
130AR70110 -- AI-generated
AI-generated
130VI753692
130VI753692 -- Client photo
Client photo
130VI753692 -- AI-generated
AI-generated
432AV212375
432AV212375 -- Client photo
Client photo
432AV212375 -- AI-generated
AI-generated
501AN0029408
501AN0029408 -- Client photo
Client photo
501AN0029408 -- AI-generated
AI-generated
130JM1003501
130JM1003501 -- Client photo
Client photo
130JM1003501 -- AI-generated
AI-generated
130JM05212510
130JM05212510 -- Client photo
Client photo
130JM05212510 -- AI-generated
AI-generated
034CC5031011
034CC5031011 -- Client photo
Client photo
034CC5031011 -- AI-generated
AI-generated
139AV351830
139AV351830 -- Client photo
Client photo
139AV351830 -- AI-generated
AI-generated
432AV2451100
432AV2451100 -- Client photo
Client photo
432AV2451100 -- AI-generated
AI-generated

Pendants & Necklaces

016SB1141
016SB1141 -- Client photo
Client photo
016SB1141 -- AI-generated
AI-generated
016JM--46
016JM--46 -- Client photo
Client photo
016JM--46 -- AI-generated
AI-generated
366JY5287
366JY5287 -- Client photo
Client photo
366JY5287 -- AI-generated
AI-generated
154JM4003
154JM4003 -- Client photo
Client photo
154JM4003 -- AI-generated
AI-generated
019DM-0251
019DM-0251 -- Client photo
Client photo
019DM-0251 -- AI-generated
AI-generated

Bracelets

060JM300107
060JM300107 -- Client photo
Client photo
060JM300107 -- AI-generated
AI-generated
012LS05040710
012LS05040710 -- Client photo
Client photo
012LS05040710 -- AI-generated
AI-generated
060JM2841
060JM2841 -- Client photo
Client photo
060JM2841 -- AI-generated
AI-generated

Inside the pipeline

This isn’t one model and a lucky prompt. It’s a production system built from several specialized parts, each doing one job well, the way a real studio splits lighting, styling, and retouching across different people instead of asking one person to do everything.

Reading the reference before generating anything

A dedicated vision model studies every reference photo first, and only that description, not the raw photo, reaches the generation step. Splitting “look at it” from “build it” produces more accurate jewelry-specific detail (exact prong counts, stone-setting styles, chain rhythms) than asking one model to see and draw in the same breath.

The right generator for the right shot

No single image model is best at everything. Clean studio isolates, worn lifestyle shots, and the fine material rendering a diamond or a polished band needs each pull from a different model, chosen for that job, not defaulted to whichever one is convenient.

A material library, not a guess

Gold doesn’t reflect light like silver, and silver doesn’t reflect light like platinum. Each metal has its own real specular behavior, tint, and edge sharpness that a generic “shiny metal” prompt gets wrong by default. Every metal in the catalog is backed by its own researched rendering language, checked against real photography.

A stone library, not a sparkle filter

A diamond’s brilliance, fire, and scintillation are three different optical effects, and they’re different again for a sapphire, a pearl, or an opal. Every stone type has documented optical behavior on file, so a render reads as a real gem under real light instead of a generic sparkle painted on top.

A pose and wardrobe library, tracked, not repeated

Worn shots draw from a structured library of hand poses and wardrobe options, split by register and gender, with the last several pieces excluded from the draw so the catalog doesn’t quietly repeat the same pose. A real catalog looks like many different moments, not one photo reused with a different ring dropped in.

Purpose-built 3D, when 2D isn’t enough

Some angles fight a purely 2D generation every time: a true edge-on profile, an exact hero tilt. For those, a small custom geometry model locks the camera angle by real math instead of hoping a prompt gets read correctly. Built once per category, reused across every piece in it.

Delivery-grade, not generation-grade

A generated image at delivery resolution isn’t the same as a generated image, period. Every approved shot goes through its own upscale and compression pass tuned for catalog and web use, so what ships is print-sharp, not just screen-preview sharp.

Seven distinct systems, one consistent shot. That’s the gap between a person typing a prompt and a pipeline built to run a catalog.

A living library

This is a living library, not a finished set. New pieces get added as they’re approved, and the shot list above is the standard applied to every one of them, today and going forward.

· START HERE

Want something like this made?

A message straight to me, usually answered the same day.
Message on WhatsApp →
+216 27 065 645