Product · Review & QA

Automatic QA on every image, at any volume.

Turn it on, and the AI reviewer checks each result against your model, your prompt, and every garment in the look. Pass, or retry with the reason attached — no hand-reviewing required.

Model wearing a cropped white jacket with relaxed denim jeans, approved by QA
Approved

result #58976

10 of 10 passed
Avatar consistencyPrompt match

Cropped jacket

FitColorsLengthDetails

Denim jeans

FitColorsLengthDetails

Approved into your library.

The question set

Every image answers the same questions.

The AI reviewer compares each output to the inputs it was generated from. With multiple garments in a look, each one is reviewed on its own.

The model

Is this the same model?

Face, hair, and features are compared to the avatar reference.

Avatar consistency

The prompt

Is this the shot that was asked for?

Pose, framing, and scene are compared to the final prompt.

Prompt match

Each garment

Is the product true to its reference?

Every garment in the look is checked against its own reference image.

FitColorsLengthDetails

The loop

Rejected work retries itself.

A failed check comes back with a reason. The retry carries it into the next attempt, and the approved take replaces the rejected one. This is the same look from the review above, retried and approved.

First attempt at the jacket and culottes look, rejected by QARejected
Details failed: five visible buttons instead of the four on the reference garment.
Retry with the reason
Retried jacket and culottes look, approved by QAApproved
All checks passed. The approved take ships; every attempt stays on the record.

How you run it

One reviewer, four ways in.

Toggle QA on any run, or run it on anything after the fact. Studio, the agent, the API, and the MCP all reach the same reviewer.

01 · Studio

Run it in the interface.

Flip QA on when you generate, or open any result and run it on the spot. The verdict lands right on the image.

Explore Studio
02 · the agent

Ask Mouza to review it.

Mouza queues QA, reads the verdicts, and retries failures with the reason attached. This exchange is real.

Explore the agent
MMouzareviewing
Run QA on the culottes look.
Reviewed. 1 of 10 checks failed: the jacket shows five buttons, the reference has four. Want me to retry with the fix?
Retry it.
Retried and approved. All 10 checks pass.
03 · the API

One flag on the run.

Turn QA on per job and read the structured verdict back on every result, reason included.

Explore the API
curl https://api.uwear.ai/generation \
  -H "Authorization: Bearer $UWEAR_API_KEY" \
  -d '{
    "clothing_item_id": 18967,
    "avatar_id": 290,
    "model_slug": "nano-banana-2",
    "do_qa": true,
    "max_qa_retries": 1
  }'

# verdict comes back on the result: qa_decision, criteria, reasons

04 · MCP

Let your agents review.

Connect Uwear as an MCP server and your agents queue reviews and read the same structured verdicts.

Explore MCP
UUwear MCPconnected
QA everything we generated today and flag what fails.
Queued QA on 3 results. 2 approved; 1 rejected on garment details: five buttons instead of four. Retrying it now.

uwear.read_generation_result_qa(58982)

Enterprise

Your reviewers' checklist, automated.

On enterprise plans, we tune the question set to your catalog and your standards. The checks your team runs by eye today become questions every image has to answer, at any volume.

Custom question setEnterprise
  • Logo crisp, unwarped, correctly placed
  • Neckline sits exactly as the flat
  • No jewelry unless the brief specifies it
  • Hosiery and footwear follow the styling guide

Example checks. Yours are written with our team from your existing review standards.

Review & QA FAQ

How automated review behaves inside Uwear production workflows.

An AI reviewer compares each output to its inputs: the model is checked for avatar consistency, the shot is checked against the final prompt, and every garment in the look is checked against its own reference for fit, colors, length, and details. Every answer comes back with a written reason.

Turn it on at the generation level so every result is reviewed as it lands, run it on demand from Studio, or queue it through the API and MCP. Agents can queue QA and read verdicts as part of their own workflows.

No. A separate AI reviewer assesses whether the inputs match the output, so it scales with your volume and never gets tired. Your team still reviews in Studio on top of it; QA just makes sure the obvious failures never reach them.

The verdict is stored with the failing check and its reason. Retry the image with that reason attached, and the approved take replaces the rejected one. You can also allow automatic QA-triggered retries per run.

Yes, on enterprise plans. We tune the AI reviewer’s question set to your catalog and standards, so the checks your team runs by eye today become questions every image has to answer.

Put a reviewer on every run.

Bring one catalog workflow. We'll wire QA into it, with your questions.

Existing customer? Log in to Studio

White cropped jacket and washed-blue trousers against the Noir Chrome wall
Black velvet-floral midi dress in profile under the Noir Chrome rim light
Brown strapless mini dress beside a chrome chair on the glossy black floor