Illustrations on Seeing Words are generated by OpenAI's GPT Image 2 image model, accessed through the KIE generation platform. If that model is unavailable, generation falls back to Google's Nano Banana 2 image model. Seeing Words does not train models on your content and does not present the model provider's name or logo as our own brand. Generated illustrations are marked as AI-assisted in the editor before you use them and are never presented as human-made artwork.

Seeing Words prohibits generating or uploading the following seven categories of content: sexual or NSFW content; violent or gory content; hate speech; child sexual abuse material (CSAM) or any content that sexualizes minors; deepfakes or content that impersonates a real person; non-consensual intimate imagery (NCII), including sharing, editing, or fabricating intimate images without the depicted person's consent; and content that infringes copyright or trademark.

Zero tolerance, immediate action. We remove the content and may report it to the authorities the moment we learn of: child sexual abuse material or any sexual content involving minors; content that promotes or assists terrorism, extremism, or mass violence; instructions for building weapons of mass destruction; content that incites genocide or ethnic hatred; and deepfakes of a real person, including face swaps, undressing, or intimate-scene composites, whether or not that person consented.

Restricted categories, allowed only under conditions: violent content only where there is a clear creative or educational context; politically sensitive content that could mislead; medical, legal, or financial advice only where it satisfies the rules of the relevant jurisdiction. Seeing Words does not offer adult content, so adult material has no permitted tier here.

Prohibited conduct: do not generate or spread disinformation; do not impersonate a real person, brand, or institution; do not infringe intellectual property, portrait, or privacy rights; do not present AI content as human-made; do not mass-produce spam or fraudulent content; and do not use jailbreak prompts to bypass our safety checks.

Uploaded photographs and voice: some templates let you upload photographs or a voice note. Upload only your own material, or material you have the explicit consent of every person shown or heard to use. Uploading another person's photograph or voice without their consent is prohibited, and the upload screen asks you to confirm that the material is yours or authorized. Consent never makes the deepfake content described above acceptable.

Automatic screening before generation: every image-generation prompt is sent to the Waffo Prompt Sift content-safety API and screened before it reaches an image model. A prompt that is blocked, or that is held for review, is never sent to a model, and we tell you that the request does not meet this Policy. If the screening service cannot be reached, we stop rather than generate unchecked.

Input validation and human review: every prompt also passes structural validation on our side; control characters, bidirectional text markers, and over-length input are rejected before generation. Content that screening flags, and content that users report, is reviewed by our team through the trust and safety address below. Reviewers are trained on this Policy and keep reports confidential.

Moderation records and severity levels: we retain screening verdicts and your reports, and we review them to check that our automatic screening is working as intended. Levels: L1 Critical (child safety, terrorism) blocked automatically, confirmed by a human, immediate ban and report to authorities; L2 High risk flagged automatically, reviewed on priority, content removed and account suspended; L3 Medium risk human review, warning and removal; L4 Low risk opened by a report, warning and required correction.

You can report content or behaviour that breaks this Policy, including child-safety or terrorism content, and AI-generated content that impersonates you or infringes your rights.

We keep a reporter's identity confidential, review every report independently, and record and act on reports that are deliberately false.

L1 Critical: first response within 2 hours, resolved within 24 hours. L2 High risk: first response within 24 hours, resolved within 3 business days. L3 Medium risk: first response within 3 business days, resolved within 7 business days. L4 Low risk: first response within 5 business days, resolved within 15 business days. When a case needs longer we tell the reporter and explain why. Every report follows the same path: received, triaged, reviewed, decided, enforced, and the outcome notified.

We may issue a written warning, remove the content, reduce generation quotas or model access, suspend the account pending review, permanently close the account, or report to law enforcement and cooperate with an investigation. We apply the lightest measure that fits the violation, then escalate for repeated or serious violations. Every measure listed here is one we can carry out.

If you believe we applied an enforcement measure in error, write to appeals@seeingwords.love within 30 days of the notice and include the card link or reference. A reviewer who took no part in the original decision looks at the case again, and we finish that review within 10 business days.

We publish a content-safety transparency summary once a year, and we give at least 15 days notice before a material change to this Policy. Continuing to use Seeing Words after an update means you accept the updated Policy. This document is a content policy, not legal advice.

Content safety and reports: safety@seeingwords.love. Appeals: appeals@seeingwords.love. General support and privacy: hello@seeingwords.love. Seeing Words operates the digital greeting-card service at seeingwords.love.