Skip to content

Photoroom launches Visual Agents to generate, check and fix product visuals, so enterprises no longer have to build their own AI stack

Visual Agents builds the right workflow for each catalog, then catches and fixes flawed images and video before they reach buyers.

Photoroom today launched Visual Agents, a managed system that runs AI image and video production for enterprise product catalogs. It arrives with internal benchmark results showing that the best AI image-editing models on the market pass a full product fidelity check only 29% of the time, and distort or render logos unreadable in one in five outputs. Photoroom tested 4,250 AI-generated images across 850 real products spanning clothing, footwear, bags, jewelry and accessories. Its own editing stack, which runs as the first step inside Visual Agents, raised that pass rate to 38.2%, the highest recorded in the test.

That gap is why several of the world's largest marketplaces have built their own in-house image scoring systems. Until now, no commercial option existed.

The enterprise challenge in AI visual production is no longer capability. It is assembly. Teams have to work out which tools solve their visual problems, then stitch multiple models and vendors together to cover what one tool cannot do alone. What that leaves behind is inconsistent output, quality review that stays manual, and custom workflows that cost more to maintain than they did to build.

How it works

Visual Agents removes that work. The system reads each incoming visual, selects the models best suited to it, scores the finished result against the product it is meant to show, and automatically retries anything that misses, sending only genuinely borderline cases to a human reviewer.

Visual QA: analyze every input. Reads each image or video against your criteria to understand what it needs and route it to the right workflow.

Create & Transform: generate and edit. Uses the AI models best suited to each input, from background removal and relighting to ghost mannequins and on-model visuals.

Fidelity Raters: score the output. Scores each output against your quality threshold, sending approved visuals forward and anything below it to Visual Fix.

Visual Fix: retried until it passes. Retries failed outputs with adjusted parameters or different models, while isolating cases that need human review.

Why the loop repeats

In testing, Photoroom's engineers found that the same garment could fail a fidelity check on one pass and clear it on the next, with no change to the model in between. Retrying automatically recovers a share of those failures before a reviewer sees them, so human review is reserved for the cases that genuinely need it.

Early enterprise deployments

  • One of the world's largest consumer marketplaces, which ran more than 1.5 million automated quality checks in three weeks and corrected more than 90,000 images where a check found a defect.

  • A food delivery marketplace that processed 8,200 catalog images in a single overnight run, with 99.9% completing automatically and seven routed to a human reviewer.

  • A global fashion resale marketplace that measured a 2.2% lift in gross merchandise value over seven days against a control group, ranging from 2% to 5% by product category, using Photoroom's Create & Transform API, which now also runs inside Visual Agents.

Customer results are anonymized at the customers' request. Named case studies will follow in September.

"I have run imaging operations that processed millions of products a year. The part nobody sees is the review queue. Every image that needs a person is a minute of somebody's day, and at catalog scale those minutes are headcount. Visual Agents cuts that queue down to the images that genuinely need someone."

Jeff Strauss, Head of Imaging at Photoroom, previously at Gap, Amazon and Wolt

"Almost every enterprise team we talk to describes the same problem. They have good models available to them, and they still cannot get a catalog they trust out the other end. That is not a model problem. It is everything around the model: knowing which one to use for which product, checking whether the result is good enough to sell, and fixing it when it is not. That work has been landing on our customers. We built Visual Agents so it lands on us instead."

Matt Rouif, Co-founder and CEO of Photoroom

Availability

Visual Agents is now available to enterprise customers, with pricing based on volume. Enterprises with stricter fidelity requirements can add the Enterprise Guarantee, the contractual commitment to output quality. The agent runs its own retry loop first, and the guarantee covers what still misses.

Talk to our enterprise team about a workflow for your catalog.

Lianne HunterLianne leads Photoroom's PR efforts, connecting the company's story with journalists, publications, and the wider media world.
Photoroom launches Visual Agents to generate, check and fix product visuals, so enterprises no longer have to build their own AI stack

Keep reading

Photoroom partners with Genesis Cloud to lower carbon emissions
Lauren Sudworth
Picking a state management library for a React app used by millions (and why we went with MobX)
Eliot Andres
Playing to win: the unexpected way we innovate at Photoroom
Matthieu Rouif
Post-mortem: Photoroom API service degradation – October 11th, 2024
Eliot Andres
Redesigning an Android UI used by millions
Aurelien Hubert
Sell from anywhere: Photoroom for Shopify comes to Android
Lianne Hunter
Senior Claude Reviewer Is Not a Good Use of Engineering Talent
Marek Zaremba-Pike
Why sequential testing is the right way to experiment at the speed of PLG apps like Photoroom
Charlotte de Thiersant
So you want to rent an NVIDIA H100 cluster? 2024 Consumer Guide
Eliot Andres
Make stable diffusion up to 100% faster with Memory Efficient Attention
Matthieu Toulemont