AI content safety & reporting
How we make illustrations, check them, and handle concerns.
Pebblehug is designed for adults creating gentle family storybooks. Our Acceptable Use Policy explains photo permissions, prohibited uses and the actions we can take.
Pebblehug is operated by geng kang, an individual based in the United States.
Contact: Pebblehug Support · support@pebblehug.com
How illustrations are made
Our current illustration integration uses OpenAI GPT Image 2.5 Flare, accessed through OpenRouter. The request selects the OpenAI provider, high image quality, and no fallback to another provider. Pebblehug combines prepared stories and illustration instructions with AI-generated images and automated content checks; it is an independent product, not endorsed by either provider.
For automatic character cropping and map placement, we also use openai/gpt-4.1-mini through OpenRouter with the OpenAI provider. It analyzes generated artwork; automated checks may miss visual mistakes.
When you upload photos, the same model finds the people in each photo so we can suggest a crop and photo tips. It does not identify who anyone is. How automatic photo checks work
Model source and version details
Configured API identifier: openai/gpt-image-2.5-flare.
On 2026-10-03, OpenRouter’s public provider catalog listed OpenAI with endpoint OpenAI | openai/gpt-image-2.5-flare-20260908. OpenAI’s model documentation lists the dated snapshot gpt-image-2.5-flare-2026-09-08.
Our request uses the undated model alias. The catalog and published snapshot have matching dates, but the alias is not pinned to that snapshot and may change. This check identifies the currently listed source; it does not prove a fixed version for every past or future image.
OpenRouter model listingProvider catalogOpenAI model documentation
Checks before content is shown
- Before generation: we check submitted details, the actual generation instructions, uploaded photos, the photo crops you confirm and illustration references.
- Before a cover appears: the generated cover stays private until its image check passes. A pending check, detector error or unavailable service keeps the image unavailable. Continuing a pending check reuses the saved image.
- Before preview pages appear: each preview spread stays private until its own checks pass, and the words of your dedication are checked when you save them. A page that doesn’t pass is not shown; it is made with the rest of the book, and the cover can still be ordered.
- Before book delivery: we check generated pages and the final images after editing and typesetting. Complete personalized books are delivered automatically when content safety and file checks pass.
- When content is accessed: approval must match the current safety policy and the actual image. Withdrawing approval prevents future access to affected images and the book PDF.
What automated checks can and cannot do
Our moderation integration uses SeeAPI. Its nsfw-filter/image-moderation checks images for NSFW and special-care flags. Its text-nsfw-filter/text-moderation checks text for explicit or suggestive sexual content, sexual content involving minors, violence, hate, harassment and self-harm.
Because that detector often flags ordinary family photos and pages, anything it blocks automatically gets one second look from an OpenAI model (openai/gpt-4.1-mini), sent through OpenRouter in the same way as our illustration requests. The item is released only if that review is confident it is safe; if it finds a problem, can’t tell, or isn’t available, the item stays blocked and we ask you to choose a different photo or request a review. Blocks placed by our operators are never released this way.
The image detector does not cover every violent, hateful, deceptive or infringing image. Automated checks can miss issues and can flag ordinary images incorrectly. A flag is a reason to review content, not a verified allegation about a person. Human review and reports remain necessary; passing a detector does not guarantee safety, accuracy or ownership of rights.
Generation and moderation providers receive the material needed for their work. Our storage is private, and image moderation uses access links that expire after 15 minutes. Under their published terms, these providers don’t use your photos or generated images to train AI models. Illustration requests go through OpenRouter, which doesn’t store request content, and are limited to OpenAI with OpenRouter’s data-collection opt-out. OpenAI may keep request content in abuse-monitoring logs for up to 30 days, or longer where the law requires. SeeAPI doesn’t publish a retention period. Provider copies follow each provider’s own deletion schedule; we can’t erase them on request. See our Privacy page.
Report a concern or request review
No account is required. This form covers safety, privacy, consent, copyright and impersonation concerns, and requests to review a blocked result. Reports are saved for authorized operators; a confirmation reference means the report was recorded, not that the review is complete.
For an email report or follow-up, include your report reference if available. Contact: Pebblehug Support · support@pebblehug.com. Do not send children’s photos, intimate material, passwords or private access tokens. Email does not create a report-form confirmation reference.
What happens after you submit
- Keep your report reference. An authorized operator reviews the concern and the relevant available records.
- If action is needed, the operator can withhold content, withdraw image approval or request another automated check, and records the reason and outcome. Content is not released just because a review was requested.
- If you provided an adult email, we can ask for clarification and reply about the outcome without disclosing another family’s private data. Anonymous reports can still be investigated.
For payment, delivery or refund issues, use the support process in our terms. A content report does not itself cancel an order or issue a refund.