Privacy

What a fusion stores

Your library picks, typed text, and any upload's written note are stored with the result so the fusion stays replayable: the brief, scores, rarity, image prompts, seeds, model and image-provider details, and an automated quality-review summary.

Only images that pass the automated quality gate are saved in the public generated-image directory and shown with the fusion. Images that fail review are not saved. Up to 200 recent mashes are kept on the demo server; older idea records are dropped automatically.

There are no accounts, tracking cookies, or analytics scripts in this demo.

Uploaded references in product fusions

Product-reference uploads must show non-human objects or materials only; images containing any person are refused. The image is re-encoded in memory to apply orientation and remove source metadata, then sent to configured vision models to screen for people and unsafe content and produce a short object-only design caption. If screening fails or is uncertain, the mash is refused. The original image bytes are not included in stored ideas and are not kept on Mapper's disk.

Configured AI providers process the image bytes for safety screening and captioning. Their retention and training practices are governed by their terms; Mapper cannot promise deletion of provider-side data. Do not upload sensitive information or an image you lack permission to use.

Family comic portraits (disabled by default)

This feature is disabled unless a deployment explicitly enables it. When available, it requires an explicit attestation of permission for every pictured child and adult. A JPEG, PNG, or WebP photo of up to 6 MB is decoded and re-encoded in memory to apply orientation and remove source metadata such as EXIF before configured AI providers receive it for safety analysis and generation.

The feature proceeds only when the classifier reports visible people, a safe-to-transform image, and no uncertain apparent ages. Automated age and safety classification can be wrong and is not age verification. The flow uses broad visual descriptors for comic-ink artwork only—never photorealism, future-child prediction, identity recognition, or an exact likeness. The generated portrait is reviewed and returned temporarily; neither source photo nor portrait is saved to Mapper storage or mash history.

Configured third-party providers process the normalized source photo and derived descriptors under their own terms. Fusion cannot promise provider-side deletion. Unclear classification, unsupported content, provider outage, generation failure, or review failure prevents a portrait from being returned.

Rate limiting

Fusions and enabled family-portrait requests share the demo's visitor/global daily quotas and concurrency controls. The quota ledger stores salted SHA-256 hashes of visitor addresses rather than raw addresses. The ledger resets daily. In-process locking is not a distributed multi-instance guarantee.

Generated content and review

Briefs and concept images are AI-generated. An automated reviewer scores brief specificity, input fusion, buildability, image clarity, coherence, brief fit, composition, artifact control, and style compliance. Weak image candidates may be regenerated once; candidates that do not pass are withheld. Review is fallible and is not a guarantee that content is safe, original, accurate, or suitable for a campaign.

Concept images are illustrations, not proof of a working product, a real person, a brand relationship, or a verified claim. Check outputs yourself before sharing, manufacturing, or using them in a crowdfunding campaign.

Contact

Questions, takedowns or abuse reports: support@navigatorslab.com. This demo is operated by the Navigators Lab team.