Human-Edited Portrait Retouching Dataset – 6 Before/After Pairs for AI
Generative AI & Computer Vision
Tags and Keywords
Free
About
Human-Edited Portrait Retouching Dataset
A professionally prepared image-editing dataset containing 6 paired source → target portrait examples created by human retouchers.
Each source image is paired with a professionally retouched target created manually in Adobe Photoshop without AI-generated editing tools.
The dataset demonstrates how human image-editing specialists can produce structured visual ground truth for training, fine-tuning, evaluation, and benchmarking of image-editing and generative AI models.
Unlike basic image annotation datasets, this dataset contains professionally corrected target images showing the intended final visual result.
Data Product Features
Each dataset record includes:
- Source Image — original portrait before editing.
- Target Image — professionally human-edited version of the same image.
- Editing Instruction — natural-language description of the requested transformation.
- Edit Summary — concise explanation of the changes performed.
- Edit Operations — structured list of individual retouching operations.
- Preserved Attributes — elements that should remain unchanged during editing.
- Technical Metadata — information about task type, editing method, software, source/target relationship, and provenance status.
The represented editing operations include:
- Skin cleanup
- Temporary blemish removal
- Redness reduction
- Skin-tone balancing
- Skin-texture refinement
- Under-eye correction
- Eye enhancement
- Lip refinement
- Local facial tonal correction
- Beauty retouching
- Male portrait retouching
- Retouching under uneven or difficult lighting
The exact editing operations vary by record.
Distribution
The dataset is delivered as a downloadable ZIP package.
Data Volume:
- 6 source images
- 6 corresponding target images
- 6 source → target pairs
- 12 JPEG image files
- 6 detailed annotation records
- 6 technical metadata records
Included files:
annotations.jsonl— detailed record-level editing annotationsmetadata.jsonl— technical metadatamanifest.csv— dataset manifestschema.json— annotation and metadata field definitionsdataset_card.md— dataset documentationREADME.md— package structure and usage informationRIGHTS_AND_PROVENANCE.md— provenance and licensing information
Primary formats: JPEG, JSONL, JSON, CSV, Markdown
Usage
This dataset can be used for a range of AI and machine-learning workflows:
-
Image-to-Image Training: Train models to transform original portraits into professionally retouched outputs.
-
Supervised Fine-Tuning: Use paired human-edited examples as target outputs for image-editing model fine-tuning.
-
Instruction-Based Image Editing: Train or evaluate systems that receive an image and natural-language editing instruction and generate the corresponding edited result.
-
Generative Image Editing: Evaluate how accurately generative models reproduce professional portrait-retouching decisions.
-
Model Evaluation: Compare AI-generated edits with professionally retouched human reference images.
-
Human Ground Truth: Use professionally edited target images as reference outputs for quality evaluation.
-
Human-in-the-Loop Systems: Support workflows in which AI-generated edits are reviewed or corrected by professional image editors.
-
Visual Quality Assessment: Study realistic skin treatment, facial-feature preservation, identity consistency, and editing quality.
Coverage
The dataset focuses on professional portrait and facial retouching.
The sample includes different subjects, lighting conditions, skin characteristics, and levels of required correction.
Depending on the individual image, editing may affect areas such as skin, eyes, lips, facial tone, shadows, and localized imperfections.
Key attributes are intentionally preserved wherever applicable, including:
- Facial identity
- Facial proportions
- Expression
- Pose
- Hairstyle
- Clothing
- Background
- Composition
The goal is to modify only the requested visual characteristics while avoiding unnecessary changes to unrelated image content.
Production Method
- Editing Method: Professional human retouching
- Software: Adobe Photoshop
- AI Used to Create Target Images: No
- Dataset Structure: Paired source → target images
- Primary Task: Portrait retouching / image-to-image editing
- Annotation Language: English
Custom Dataset Production
This dataset is a small demonstration of the visual training data that FixThePhoto can produce at larger scale.
Custom datasets can be created according to buyer specifications, including:
- Human-edited source → target pairs
- Source → instruction → target data
- Multi-step editing sequences
- Portrait and beauty retouching
- Body editing
- Product and e-commerce editing
- Object addition and removal
- Background editing
- Photo restoration
- Masks and matting
- Layered PSD files
- Human correction of AI-generated outputs
- Custom edit taxonomies
- Buyer-specific metadata and QA requirements
Dataset size, editing criteria, annotation structure, file format, QA rules, and delivery workflow can be customized for individual AI and machine-learning projects.
Licensing
This dataset is provided as a demonstration sample.
The original source images originate from third-party stock sources. Source-level licensing and redistribution rights should be verified before any use beyond the applicable dataset and AI-training license terms.
Redistribution, resale, or sublicensing of the source or target image files is not permitted unless explicitly authorized.
TRUSTLoading...
Free
Download Dataset in ZIP Format
Recommended Datasets
Loading recommendations...