Our approach · Updated 20 October 2026
Sources & licences
Our skin appearance guide draws on openly available research about photographs and their limits. Here is what we use, how we adapt it, and what our estimates can tell you.
A guide for appearance, with limits
Skinreo sends a shared assessment guide with the photos you submit to OpenAI GPT-6 Luna. Both individual scans and photo comparisons use the same definitions. A three-angle scan is assessed as one set of views. We use these sources to write instructions; we have not fine-tuned the model on these articles or datasets.
The guide considers visible blemish-like features, red or pink variation, dry-looking surface changes, texture and lines at rest. A photo cannot measure skin moisture, diagnose a condition or establish why something looks different. Normal pores, freckles, facial hair and reflections are not automatically signs of a concern.
Our numerical anchors and equal-weight summary are Skinreo product choices. They are not clinically validated scales, health measurements or attractiveness ratings. A score of 77 does not establish one-point accuracy. If a feature cannot be assessed, we leave its score unavailable.
Lighting, angle, expression, image detail and visible coverage can limit comparisons. Differences between independently saved estimates do not establish a change in skin. Different scoring-guide versions are kept separate in trends.
Research used in the guide
Photography in dermatology
Zarqa Ali and Simon Francis Thomsen. Frontiers in Medicine 13:1854982 (2026).
Informs checks for lighting, focus, colour balance and positioning. We paraphrase selected guidance; figures and patient photographs are not reproduced.
© 2026 Ali and Thomsen. Licensed under CC BY 4.0. Text has been paraphrased, shortened and adapted for this cosmetic journal. Source material is provided without warranties under the licence disclaimer.
Standardized clinical photography considerations in patients across skin tones
Yuna Oh, Alina Markova, Sarah J. Noor and Veronica Rotemberg. British Journal of Dermatology 186(2):352–354 (2022), doi:10.1111/bjd.20766.
Informs caution about colour and surface visibility across skin tones and photography conditions. We paraphrase the text; clinical images are not reproduced.
© 2022 British Association of Dermatologists. The moral rights of the named authors have been asserted. Licensed under CC BY 4.0. Text has been paraphrased, shortened and adapted for this cosmetic journal. Source material is provided without warranties under the licence disclaimer.
Review of Modern Techniques for the Assessment of Skin Hydration
M. Qassem and P. A. Kyriacou. Cosmetics 6(1):19 (2019), doi:10.3390/cosmetics6010019.
Informs the distinction between visible dry-looking surface changes and actual skin hydration measurement. We summarise that limitation; no figures are reproduced.
© 2019 the authors. Licensed under CC BY 4.0. Text has been paraphrased, shortened and adapted for this cosmetic journal. Source material is provided without warranties under the licence disclaimer.
Acne — NIAMS
National Institute of Arthritis and Musculoskeletal and Skin Diseases, National Institutes of Health. Reviewed July 2023.
Used as background vocabulary for visible spots and bumps, not to request medical classifications. NIAMS identifies its website text as public domain. We paraphrase the text; we do not reproduce photographs or illustrations from that page.
SCIN: considered for future evaluation
The Skin Condition Image Network (SCIN) dataset from Google Research and collaborators contains consented dermatology images and condition labels. It is a possible source of test images across photography conditions and skin tones. Its labels are not reference values for Skinreo’s cosmetic scores or proof of before-and-after changes.
SCIN has not been used to fine-tune or evaluate Skinreo’s model. Its images are not included in routine scan requests or reproduced here. The SCIN Data Use License grants royalty-free reuse subject to attribution, notice retention, a prohibition on re-identification and its other conditions. It does not generally grant privacy or publicity rights. Any future use must follow those conditions, including when passing material to an AI provider.
Our scoring anchors
Guide: skinreo-appearance-v1. These are our provisional descriptions. Intermediate integers are allowed when supported by visible evidence. Each result includes the closest appearance anchor, and we check that its score follows the correct direction. The provisional bands are: no feature apparent = 0, subtle = 1–37, clear = 38–62, pronounced = 63–87, extensive = 88–100; the visible-dryness score reverses that direction. The assessed facial regions are recorded, and hidden skin is not treated as clear skin.
| Appearance feature | 0 | 25 | 50 | 75 | 100 |
|---|---|---|---|---|---|
| Blemishes | None apparent | A few subtle spots or bumps | Multiple clear features | Many prominent features across regions | Extensive, densely distributed features |
| Redness | No distinct variation apparent | Subtle localised variation | Clear localised or moderate multi-area variation | Pronounced variation across regions | Extensive, very pronounced variation |
| Surface texture | No distinct unevenness apparent | Subtle localised unevenness | Clear uneven surface | Prominent unevenness across regions | Extensive pronounced unevenness |
| Lines at rest | None apparent | A few faint lines | Clear fine lines in multiple areas | Several pronounced lines | Extensive prominent lines |
| Visible dryness (inverse scale) | Extensive pronounced dry-looking changes | Pronounced patches across regions | Several patches or localised flaking | Subtle localised changes | No dry-looking changes apparent |
The internal field called “hydration” is an inverse visible-dryness estimate: higher means fewer apparent dry-looking changes. Shine alone does not establish hydration. Earlier scans used a different dryness/dewiness guide and remain labelled accordingly.
When all five components are assessable, the overall appearance summary is the rounded average of: 100 minus blemishes, 100 minus redness, the inverse dryness score, 100 minus texture, and 100 minus lines. All five have equal weight. Each feature uses its relevant visible regions; comparisons over time still require matching regions for each feature. We leave the summary unavailable if a component cannot be assessed.
Licences and independence
CC BY 4.0 permits commercial reuse with its attribution and other conditions. The licence applies to the identified source material; it does not mean that Skinreo’s app or OpenAI’s model is open-source software. Public access alone does not give permission to reuse every image or figure.
Third-party material retains its own licence. Skinreo’s general restrictions on copying the service do not override rights granted by those licences. We preserve source credits in the guide sent to our AI provider. No patient images or figures from these publications are used in the guide.
Source authors, institutions, publishers, Google and NIH do not endorse, sponsor or validate Skinreo. We do not use their logos to suggest an affiliation. We do not use DermNet text or photographs in this assessment guide.
For personal-data processing, read our Privacy Policy. For questions about a source or credit, contact us.
Apple purchase verification
Our server uses Apple’s App Store Server Library for Node.js to verify App Store purchases. Copyright © 2023 Apple Inc. The library is available under the MIT License; read its copyright and licence notice. Apple’s public root certificates verify purchase signatures. These software credits do not imply Apple endorses Skinreo’s appearance estimates.
Image usage and permissions
Image credit: Skinreo. These illustrations accompany our blog articles.
For questions about reusing an illustration, email hello@skinreo.com with the image or article URL and your intended use, or contact us. We will confirm any applicable rights and terms.
Any permissions we offer apply only to rights we hold. The research licences above apply to the identified source material. Uses permitted by law and separately identified third-party image licences remain unaffected.