AI Transparency Notice

Last updated: August 6, 2026

In short: every image and every audio story you receive in She Affirms is generated by artificial intelligence. Nothing you see or hear is a photograph, a recording of a real person, or a depiction of a real event. The affirmation text is AI-written too. You are never talking to a human being.

This notice explains which AI systems She Affirms uses, what they do with your input, how generated content is marked, and what we deliberately do not do. It is published to satisfy Article 50 of Regulation (EU) 2024/1689 (the EU AI Act), which has applied since 2 August 2026, and it complements our Privacy Policy, Terms of Service, and Sub-processors list.

1. Our role under the AI Act

Jasper Slowik, operating She Affirms (see Impressum), is the provider of the AI system "She Affirms" within the meaning of Art. 3(3) AI Act: we place a generative AI system on the market under our own name. At the same time we are a deployer of third-party general-purpose AI models, which we access through the APIs listed in section 2.

When you export or share a generated card outside the App — to social media, a messaging app, or anywhere else — you act as the deployer of that content and Art. 50(4) applies to you. See section 6.

2. Which AI systems we use

Component Model / provider What it does Your input
Story & affirmation text Gemini 3 Flash — Google LLC (Vertex AI) Turns your wish into a story arc, scene description, and affirmation line Your wish text and the personal context you entered during onboarding
Manifest image Nano Banana Pro — Google LLC (Vertex AI) Renders the 4:5 cover image in the style pack you chose The generated scene description and style parameters
Narrated audio story Turbo v2.5 — ElevenLabs Inc. Synthesises the 45–60 second whispered narration The generated affirmation script and voice parameters
Voice input (optional) Whisper API — OpenAI, L.L.C. Transcribes a spoken wish into text A short voice recording, deleted after transcription
Safety filtering Google Vertex AI safety filters; ElevenLabs safety classifier Blocks prohibited content before and after generation Your wish text and the generated script

None of these providers are permitted to train their models on your wishes, your voice, or your generated content. Data locations and transfer mechanisms are documented in Sub-processors.

3. You are interacting with an AI (Art. 50(1))

She Affirms has no human authors, no coaches, and no support agents writing your cards. The narrated voice is a synthetic voice, not a recording of a real narrator. Every generation screen and every generated card is labelled as AI-generated inside the App, at the latest at the moment you first see the content.

4. How generated content is marked (Art. 50(2))

Art. 50(2) requires providers of generative AI systems to mark outputs in a machine-readable format so that they are detectable as artificially generated. Our current state:

  • Images carry Google SynthID — an invisible watermark embedded by the image model and detectable by SynthID-compatible tooling. It survives common edits such as cropping, resizing, and re-compression.
  • Audio carries the ElevenLabs audio watermark applied at synthesis time.
  • Text (affirmation lines and story scripts) is not machine-readably watermarked. It is labelled visibly as AI-generated wherever it is shown.
  • C2PA Content Credentials in exported PNG and MP3 files are in progress. Art. 50(2) allows a transitional period until 2 December 2026 for AI systems placed on the market before 2 August 2026, and we are working within it.

Free exports additionally carry a visible She Affirms watermark. That watermark is branding, not the legally required marking — the SynthID and ElevenLabs marks described above remain embedded in Premium exports as well.

5. Marking on this website

The style previews and card examples shown on sheaffirms.app are AI-generated images produced with the same pipeline described above. They are labelled as such on the page. Product screenshots show the real App interface. No image on this site depicts a real, identifiable person.

6. If you share your cards (Art. 50(4))

Once you publish a generated image or audio story outside the App, you become the deployer of that content. Under Art. 50(4) AI Act you must disclose that the content is artificially generated if it could be mistaken for a real person, object, place, or event.

Our images are stylised by design and our prompts explicitly exclude recognisable celebrity faces, but the responsibility for disclosure when you post sits with you. A simple "AI-generated with She Affirms" caption is enough. Many platforms also offer an AI-content toggle when you upload — please use it.

7. What She Affirms does not do

We consider She Affirms neither a prohibited AI practice under Art. 5 nor a high-risk system under Annex III of the AI Act. Specifically, the App does not:

  • infer, recognise, or categorise emotions (Art. 50(3) does not apply);
  • perform biometric identification or biometric categorisation;
  • use subliminal, purposefully manipulative, or deceptive techniques beyond the affirmation practice you consciously choose and control;
  • exploit vulnerabilities related to age, disability, or social or economic situation;
  • perform social scoring or any form of profiling that produces legal or similarly significant effects;
  • make automated decisions about you within the meaning of Art. 22 GDPR;
  • clone your voice or synthesise speech in your likeness;
  • generate content in the areas of education, employment, credit, insurance, law enforcement, migration, or access to essential services.

She Affirms is a creative and reflective tool. As stated in our Terms of Service, it is not medical, psychological, financial, or legal advice, and no generated content should be treated as a prediction or a guarantee of any real-world outcome.

8. Accuracy, limitations, and human oversight

Generative models are probabilistic. Output can be inaccurate, odd, repetitive, off-style, or occasionally not what you asked for. Images may contain anatomical or typographic artefacts. The narrated voice may mispronounce names or uncommon words.

You stay in control: you write the wish, you choose the style and voice, you decide whether to keep, favourite, delete, or regenerate a card, and you decide whether to share it. Every generation can be deleted from your library at any time. If a generation is wrong, offensive, or unsafe, report it to contact@sheaffirms.app or through the in-App report action — see our Community Guidelines for the moderation and appeal process.

9. AI literacy (Art. 4)

Art. 4 AI Act requires providers and deployers to ensure a sufficient level of AI literacy among the people who operate the system on their behalf. She Affirms is operated by a sole trader; everyone with access to the generation pipeline is familiar with how the models are prompted, what their limitations are, what the safety filters block, and how to escalate an incident. This notice and the Community Guidelines serve the same purpose for you as a user.

10. Complaints and supervision

If you believe She Affirms does not meet its obligations under the AI Act, contact us first at contact@sheaffirms.app. We acknowledge within 5 working days and respond substantively within 30 days.

You may also complain to the German national market surveillance authority for the AI Act, the Bundesnetzagentur (Tulpenfeld 4, 53113 Bonn, Germany), which acts as the central contact and complaints body under the German AI Act Implementation Act. For matters concerning general-purpose AI models, the European AI Office at the European Commission is competent. Data-protection complaints go to the supervisory authority named in our Privacy Policy.

11. Changes to this notice

We update this page whenever we change a model, add a generation feature, or change how content is marked. The date at the top reflects the last substantive change.

Contact

contact@sheaffirms.app