TRANSPARENCY · REGULATION (EU) 2024/1689
Artificial intelligence at Chulair
Chulair runs on artificial intelligence and we say so wherever it is used. This page is the public inventory of those systems: what each one generates, which models are behind it, how we label what is synthetic so it can be told apart from a real photo, and what we do not do with your images. We publish it in compliance with Regulation (EU) 2024/1689 on Artificial Intelligence, whose article 50 (transparency) has applied since 2 August 2026.
What AI does at Chulair
| System | Model | What it produces | What it does not do |
|---|---|---|---|
| Garment recognition | Gemini (image), Google Vertex AI | From the photo of an outfit, it crops and labels each garment and generates a clean image of each one (a synthetic "packshot"). | It does not identify the person in the photo or infer anything about them: it only works with the clothes. |
| Virtual try-on | Gemini (image) or virtual-try-on-001, Google Vertex AI | Combines your selfie with the image of a garment and produces a synthetic image of how it would look on you. | It makes no decision about you. The result is a visualisation, not a photo. |
| @chulair outfits | Gemini (image) | Fully generated images of models who do not exist, dressed in garments created for the occasion. Published in the app, on Instagram, Facebook and YouTube. | They do not depict real people and are not based on user photos. |
| Outfit animation | Veo, Google Vertex AI | Short videos generated from the image of an outfit (always for the bot; for yours only if you turn on "Animate my posts"). | Optional and off by default for users. |
| Text | Gemini (text) | Outfit descriptions, translations into the app's languages, the blog entries and the notices sent in several languages. | It is not a chatbot and does not deal with matters of public interest. |
| "Your outfit for today" | Gemini (text) | A daily suggestion with your own garments according to the season, the forecast and your preferences. | It does not receive your photos: only your garments' labels (type, colour, name). |
| Automatic moderation | Cloud Vision (SafeSearch and labels), Google | Checks, before accepting a photo or video, that clothing is visible and there is no explicit content, and again before anything is published in the showroom. | No decision affecting you is based on this filter alone: a rejection only prevents that image from being uploaded, and you can retry or write to us. |
How generated content is labelled
Everything the AI generates is labelled in two ways, as article 50 requires:
- Visibly, on first exposure. The "Generated with AI" badge on the bot's posts and on published try-on results; the text "✨ AI-generated image" on Instagram and Facebook, with the platforms' native generated-content label; the synthetic-content label on YouTube; a notice burnt into Stories; and a footnote on every blog entry.
- Machine-readable, inside the file. Generated images carry the IPTC standard
DigitalSourceType = trainedAlgorithmicMediain their metadata (thumbnails and shared versions too); generated videos carry the mark in the container metadata. In addition, Google's models embed their SynthID watermark in pixels and frames.
What is real is not labelled. The photos of your clothes, your selfies and the videos you upload are yours and are not synthetic; labelling them as generated would be a false statement.
What the AI does not do
- It does not identify or recognise people, does not recognise emotions and does not infer age, gender or any other attribute of the person. The systems work with the clothes.
- It makes no automated decisions with legal or similar effects on you. The only automatic filter is moderation, and it never has the last word.
- There is no chatbot or artificial interlocutor posing as a person.
- Google does not use your images or data to train its models, under the Vertex AI terms; nor does Chulair train models of its own.
- For all of the above, none of these systems is a prohibited practice or a high-risk system under the Regulation; our obligations are the transparency ones of article 50, which are the ones this page describes.
Models and providers
All the models are Google's and are used through its API (Vertex AI): Gemini for image and text, Veo for video and Cloud Vision for moderation. Processing is transient; depending on the feature it may take place on Vertex AI's global endpoint or, for the virtual-try-on-001 model, in the us-central1 region (USA). The detail of each processing operation, its legal basis and retention periods are in the privacy policy, section 8.
Questions, complaints and reports
If you think a generated image is not labelled, that synthetic content is passing itself off as real, or that an automatic decision has harmed you, write to soporte@chulair.com (and, for abuse, abuso@chulair.com). Every post also has a report button in the app. The supervisory authority in Spain is the Spanish Agency for the Supervision of Artificial Intelligence (AESIA).
Inventory last reviewed: 2026-09-11. Reviewed at least quarterly and whenever a system is added or removed. More in the legal notice.