This page is our transparency information under Article 50 of Regulation (EU) 2024/1689 (the EU AI Act), whose transparency obligations apply from 2 August 2026. It explains, in plain terms, where artificial intelligence sits inside Kidoreka, how we mark what it produces, and what stays entirely in your hands.
1. Kira is an AI assistant
Kira is not a person. Everywhere Kira appears — the chat window in Studio, and the ✦ sparkle buttons next to individual fields — you are interacting with an artificial intelligence system, not with a member of our team or with another parent.
We give that system a name and a friendly manner because it makes the product pleasant to use, not to suggest there is a human on the other side. The chat window states this on every screen, above the conversation.
2. AI never runs by itself
Every AI feature in Kidoreka is something you press. Nothing generates in the background, nothing rewrites what you already made, and no result is ever applied automatically:
- Kira proposes; you review the proposal and choose whether to accept it.
- Nothing is saved until you press Save. Rejected suggestions are discarded and never stored.
- AI does not read across your library, does not watch how your child plays, and does not make decisions about you or your child.
3. Which AI systems we use
We do not train our own models. We call the following third-party systems, each only for the purpose listed:
- Google Gemini — pictures: story illustrations, memory-game card art, colouring pages.
- Microsoft Azure OpenAI — writing help (story drafts, rewriting a line, song lyrics, Kira's replies), and a fallback for picture generation.
- ElevenLabs — narration voices, sound effects, and music.
The narration voices you can choose are synthetic voices from a catalogue. They are not recordings of the person named, and Kidoreka has no voice-cloning feature: we cannot and do not reproduce your voice, your child's voice, or anybody else's. When you want a real voice in a story, you record it yourself — and that recording is never sent to a model.
4. How AI-generated content is marked
Article 50(2) requires the output of generative AI to be marked in a machine-readable way. Every picture, narration, sound effect, and song that Kira generates for you carries:
- Embedded metadata stating that the file was produced by a generative model, which model produced it, and when. For images this uses the standard IPTC digital source type value
trainedAlgorithmicMedia, written as XMP and EXIF; for audio it is written into the file's tags. - An invisible watermark in the pixels of pictures made with Google Gemini (SynthID), which survives resizing, re-encoding, and screenshots.
- A record in your library, so anything Kira made is flagged as such in Studio and you can always tell it apart from your own photos and recordings.
We deliberately do not stamp a visible watermark across the artwork itself. The law asks for the marking to be detectable by machines, and a badge burned into the middle of a bedtime story would spoil the thing you made without helping anyone.
One honest limitation: embedded metadata can be stripped. If you screenshot a picture or run it through a tool that discards metadata, the embedded claim is lost. That is why Gemini's in-pixel watermark, which survives that treatment, matters alongside it.
5. What is not AI
- Photos, drawings, and recordings you upload — these are yours, and they are never presented as AI output.
- Simple editing help such as fixing a typo or shortening a sentence you wrote. This is assistive editing that does not change your meaning, which Article 50(2) expressly excludes from the marking requirement.
- The games themselves — puzzles, memory grids, and the drawing tools are ordinary software, not AI.
6. Your controls
- You can turn AI off. Personalisation is optional in Studio Settings, and no AI feature ever activates by itself.
- Children never meet the AI. The Kidoreka Play app has no AI features at all: it plays back what a parent made and saved. There is no chatbot, no generation, and no model call in the children's app.
- We never train on your content. Your stories, photos, recordings, and prompts are not used to train any model, ours or a provider's.
7. Limitations you should expect
Generative models get things wrong. They invent details, mishear words, draw six-fingered hands, and occasionally produce something that simply does not fit. Kira is a starting point for you to edit, not an authority — please read what it writes before your child does. Kira is not a source of medical, educational, or psychological advice about your child.
8. Questions
Write to support@kidoreka.com — a human reads it. See also our Privacy Policy for what data is sent to which provider, and our Terms of Service for who owns AI output.