PIXEND DATA STUDIO TRAINING PAIRS / CUSTOM DATA PRODUCTIONBUILT FOR VISUAL AI
pixend/
My requests
PAIRED IMAGE DATA FOR AI TEAMS

Better image AI.
It starts with
better pairs.

Source images, precise edit instructions, and matched outputs. Data built around the behavior your model needs to learn.

Explore datasets Inspect samples
SOURCEINSTRUCTIONTARGET
Sample photos coming soon

JPG / PNG photos and matched edits will appear here when available.

A clear instruction. A matched result.Source + Instruction + Target
Matched source–target pairs Task-specific quality criteria Project-specific usage rights Structured delivery
01 / CHOOSE HOW TO SOURCE YOUR DATA

Existing datasets. Or your own brief.

Buy existing datasetsReview quantities, specifications and delivery.Commission custom dataDefine new edits, coverage and production volume.
EXISTING DATASETS

Find a collection that fits.

JPG / PNG image pairs + an Excel spreadsheet. Purchase requests are confirmed by quotation.

Inspect samples

Listings marked “Example” contain illustrative specifications and quantities. Actual stock and delivery are not yet confirmed.

Example listingPortrait

Portrait Retouching Collection

Paired portrait photos for local skin and hair editing tasks.

Example quantity
1,200 pairs
Resolution
3000 × 4000 px
Editing tasks
Blemish removal Flyaway hair cleanup Local skin-tone correction
Delivery
Within 3 business days after payment and license agreement
View dataset
Example listingProduct

Product Cleanup Collection

Paired product photos for surface cleanup and consistent product presentation.

Example quantity
800 pairs
Resolution
4000 × 4000 px
Editing tasks
Dust removal Surface cleanup Background cleanup
Delivery
Within 2 business days after payment and license agreement
View dataset
Example listingObject removal

Object Removal Collection

Source and target pairs for specified-object removal and scene reconstruction.

Example quantity
600 pairs
Resolution
3840 × 2160 px
Editing tasks
Object removal Scene reconstruction Boundary cleanup
Delivery
Within 5 business days after payment and license agreement
View dataset
CUSTOM DATA PRODUCTION

Made to your specification.

Commission new image pairs around your editing task, source images and target volume. Scope and production timelines are quoted separately.

3 collections Custom production · Quoted by scope

Published photos are selected from the sample library. Availability, quantity, pricing, and evaluation rights are confirmed for each project.

FROM EVALUATION TO PRODUCTION

Choose your next step.

Track your requests
01 / Photos + spreadsheet

Inspect a sample

Inspect published before-and-after photos and available spreadsheet samples.

Open sample lab
02 / Quotation required

Scope a pilot

Define 100 or 300 target pairs and agree on the standard before production.

Configure a pilot
03 / Custom production

Build your dataset

Discuss larger batches, your own images, and project-specific rights.

Share your requirements
02 / INSIDE THE DELIVERY

More than images.
The context to train with.

JPG / PNG photos and a spreadsheet are our standard delivery. Pair IDs connect each photo to its editing instruction and review details. CSV, JSON, TXT and MD are available on request.

Traceable pairsStable IDs, source lineage, and versioned manifests.
Reviewable standardsTask instructions, preservation criteria, and QA records.
An agreed handoffImage format, metadata schema, checksums, and license.
Download XLSX template
pair_metadata.xlsxDEFAULT FORMAT
ColumnContents
pair_idUnique pair identifier
source_fileBefore photo · JPG / PNG
target_fileAfter photo · JPG / PNG
instructionRequested editing task
preserveDetails to keep unchanged
review_statusPending / Accepted / Rework

CSV, JSON, TXT and MD available on request.

YOUR MODEL. YOUR DATA.

Your model is specific.
Your data should be, too.

Bring your source images, or define a new collection.
We scope the edits, coverage, and delivery around your task.

01Define the sourceYour private images or a new collection
02Agree on the editing standardSpecify what changes—and what stays
03Build a structured deliveryPairs, instructions, tags, and review records
03 / OUR REVIEW FRAMEWORK

Quality is a specification.

Define acceptance criteria in the pilot.
Apply them throughout production.

01

Instruction fidelity

Review the requested edit and the preservation of everything outside its scope.

02

Provenance and permitted use

Confirm source rights, production methods, and permitted uses for each project.

03

Delivery integrity

Check pair matching, technical specifications, metadata, versions, and file hashes.

BEFORE A BATCH IS ACCEPTED

Agree on what “good” means.

The pilot defines the review rubric, tolerances, rejection reasons, and rework scope.

  • Edit compliance — the requested change is present.
  • Preservation — unrelated identity, texture, and geometry stay intact.
  • Technical consistency — pair alignment, format, and resolution meet the specification.
  • Dataset integrity — define duplicate checks and source-group splits to reduce evaluation leakage.
01Define scope
02Review samples
03Approve pilot
04Produce & review
05Deliver
LICENSED FOR YOUR USE CASE

A clear license.
A clear path to production.

From evaluation to commercial training.
Agree on the use case before the first delivery.

01Evaluation

Assess technical fit under an evaluation agreement.

02Commercial Training

Define permitted training and commercial model use.

03Custom & Exclusive

Discuss exclusivity, reuse, and recurring supply.

BEFORE WE GET STARTED

Good questions.
Clear answers.

From your first evaluation
to a recurring data supply.