---
license: cc-by-nc-4.0
task_categories:
  - image-to-text
  - tabular-classification
  - text-classification
  - feature-extraction
language:
  - en
tags:
  - beauty
  - skincare
  - ingredients
  - cosmetics
  - makeup
  - machine-learning
  - computer-vision
pretty_name: "The Beauty API: 180k+ Skincare & Cosmetics Products Dataset (Sample)"
size_categories:
  - 1K<n<10K
---

# The Beauty API: 180k+ Skincare & Cosmetics Products Dataset

**The world’s most comprehensive, normalized products database for beauty-tech and market research.**

This repository contains a structured sample of **The Beauty API**. While most beauty datasets are messy scrapes, this is a production-ready asset built on a research-backed backbone, normalized for day-one integration into recommendation engines, e-commerce storefronts, and AI models.

**[Purchase Full Dataset License](https://thebeautyapi.com)**

---

## 📊 Dataset Overview (as of 17 August 2026)

- **Total Records:** 180,000+ Unique Products, their functional classifications, ingredients list and dermatological safety
- **Brand Coverage:** 20,000+ Global Brands (e.g., Paula's Choice, CeraVe, The Ordinary, Augustinus Bader)
- **Image Archive:** 18GB of high-resolution, ID-mapped imagery (Standardized white backgrounds)
- **Barcodes:** GTIN-14 codes (UPC/EAN, globally resolvable ranges only) on ~44% of products, as a per-product `gtin14` list
- **Formats:** Interchangeable **CSV** and **JSONL**

---

## 💡 Data Capabilities

### 1. Ingredients Intelligence & Safety

Every product is parsed into a rich, structured object to bypass months of data wrangling.

- **Full INCI breakdown:** Each ingredient carries its on-label name, common aliases (`other_names`), and functional roles (`functions`, e.g. "solvent," "antioxidant," "soothing"), in exact on-pack order with label position preserved.
- **Chemical identifiers:** Regulatory-grade **CAS**, **EC**, **IUPAC**, and **Ph. Eur.** names per ingredient for enterprise R&D and compliance work.
- **Dermatological Safety:** Research-backed **comedogenic ratings (0–5)**, **irritancy scores (0–5)**, and editorial **expert ratings** at the ingredient level.
- **Free-from flags:** Boolean columns for instant filtering — `contains_fragrance`, `contains_drying_alcohol`, `contains_parabens`, `contains_sulfates`, `contains_silicones`.
- **Product classification:** Product `category` and brand `origin` for segmentation.

### 2. Machine Learning & Computer Vision

A pre-labeled training set for high-accuracy visual recognition and retail audits.

- **Branding Analysis:** Train multimodal models to decode the correlation between packaging design and ingredient quality.
- **Visual Search:** Use the 18GB mapped archive to build "Point-and-Know" camera features that bypass the need for barcodes.
- **Retail Auditing:** Power automated shelf-scanners to identify 180,000+ SKUs in real-world lighting conditions.

### 3. E-commerce Enrichment (B2B)

Metadata designed to boost conversion and lower return rates for indie e-com and retail.

- **Barcode Matching:** Join the catalogue to your own inventory or a scanner app on GTIN/UPC/EAN via the `gtin14` list.
- **Page Enrichment:** Generate SEO-rich product descriptions and ingredient glossaries.
- **Trust Signals:** Implement safety badges and personalized skin-type matching engines.

---

## Technical Schema

The dataset is delivered in interchangeable **CSV** and **JSONL** formats.

**Product fields**

| Field | Type | Nullable | Description |
|---|---|---|---|
| `id` | integer | no | Stable unique product id. Matches the image filename (`<id>.jpeg`). Ids are not strictly sequential; gaps reflect products dropped during clean. |
| `brand` | string | no | Brand / manufacturer name. |
| `name` | string | no | Product name. May contain non-ASCII characters, punctuation, or unicode whitespace. |
| `image_name` | string | yes | Filename of the product image (`<id>.jpeg`), or `null`. |
| `gtin14` | array&lt;string&gt; | no | Barcodes known for this product, each as a canonical 14-digit GTIN (left-zero-padded, so UPC-A `782446192371` is `00782446192371`); codes the product primarily owns first, then ascending. Includes unit codes and codes of sibling variants of the same line (shade/scent/size). Only globally resolvable GS1 ranges (restricted in-store/coupon ranges excluded). `[]` when none known. In CSV this is the JSON-array text in one quoted column. |
| `category` | string | yes | Product category (e.g. `skincare`). |
| `origin` | string | yes | Brand country of origin, or `null`. |
| `contains_fragrance` | boolean | no | Formulation contains fragrance/parfum. |
| `contains_drying_alcohol` | boolean | no | Contains a drying (simple) alcohol. |
| `contains_parabens` | boolean | no | Contains parabens. |
| `contains_sulfates` | boolean | no | Contains sulfates. |
| `contains_silicones` | boolean | no | Contains silicones. |
| `ingredients_simple` | array&lt;string&gt; | yes | Flat ordered list of on-label ingredient names. In CSV this is a `|`-delimited string. Order is on-pack INCI order. *(CSV convenience column, derived from `ingredients[].label_name`.)* |
| `ingredients` | array&lt;object&gt; | yes | Rich ingredient records, in on-pack order. In CSV this is a JSON-encoded string inside one quoted column. Per-ingredient fields below. |

**`ingredients[]` object fields**

| Field | Type | Nullable | Description |
|---|---|---|---|
| `position` | integer | no | 0-based slot in the on-pack INCI list. |
| `label_name` | string | no | Exact on-label text for this ingredient on this product. |
| `concentration` | string | yes | Stated concentration, if printed. |
| `ingredient_id` | integer | no | Canonical ingredient id — the same chemical shares one id across products. |
| `name` | string | no | Canonical ingredient name. |
| `rating` | string | yes | Editorial rating bucket: one of `direct actives`, `supporting actives`, `sensitizers`, or `null` (unrated, the majority of ingredients). |
| `irritancy` | string | yes | Irritancy on a 0–5 scale (higher = more irritating), or `null`. A string because values come as points or ranges (`"0"`, `"3"`, `"0-3"`). |
| `comedogenicity` | string | yes | Pore-clogging potential on a 0–5 scale (higher = more comedogenic), or `null`. Same point/range string encoding as `irritancy`. |
| `functions` | array&lt;string&gt; | yes | Functional roles (e.g. `solvent`, `antioxidant`). |
| `cas_number` | string | yes | CAS registry number. |
| `ec_number` | string | yes | EC number. |
| `ph_eur_name` | string | yes | European Pharmacopoeia name. |
| `iupac_name` | string | yes | IUPAC name. |
| `other_names` | string | yes | Known aliases / synonyms. |
| `category` | string | yes | Flag category that drives the product-level `contains_*` booleans: one of `fragrance`, `drying_alcohol`, `parabens`, `sulfates`, `silicones`, or `null`. |

---

## Licensing

This sample dataset is provided under the **CC BY-NC 4.0** license for non-commercial research and evaluation.

**Commercial Access:** For production e-commerce, commercial AI training, or retail audit tools, a commercial license is required. Visit [thebeautyapi.com](https://thebeautyapi.com) for licensing details and full dataset downloads.