AI Terms.

These terms apply when you enable AI-assisted features in the Holdstill product. They do not govern third-party tools you use outside the platform. Where processing involves personal data of gallery visitors, your relationship with those individuals remains governed by your notices and contracts; our DPA describes our processor role for gallery delivery. If any conflict arises between marketing descriptions and these terms regarding AI behaviour, these terms control for compliance purposes.

AI features are disabled at account and gallery level until you explicitly enable them. You may operate your entire practice on Holdstill without ever triggering model inference. Enabling a feature constitutes instruction for us to process the minimum inputs necessary to produce on-screen suggestions. You may withdraw that instruction by disabling the feature; withdrawal stops new processing promptly and does not erase lawful logs already written for security.

There is no fully automated decision that publishes galleries, deletes assets, changes pricing, or alters legal terms without a human confirmation step in the product flows we ship. Suggestions appear as drafts or overlays you can accept, edit, or discard. You remain solely responsible for what clients ultimately see. We prohibit using the API or UI in ways intended to bypass these safeguards.

Depending on feature flags available to your plan, models may propose narrative ordering, highlight frames that appear salient under statistical heuristics, suggest chapter boundaries, or generate short textual prompts you can edit. The models are tuned for photographic storytelling workflows, not general open-domain chat. Outputs are probabilistic and may reflect common tropes in wedding and editorial photography; they may omit unusual artistic choices that define your signature.

You retain ownership of your images and metadata. Outputs are programmatically generated suggestions and, to the extent protectable, you own the selections you save into your gallery configuration. We do not claim copyright over your imagery. You grant us a limited licence to process inputs solely to provide the service during the feature's enabled window.

We do not sell your photos or use them to train third-party foundation models for their general products. Any research-grade improvement of internal models that could involve retained content beyond immediate inference is contractually constrained and, where required, opt-in with clear disclosure. We do not build advertising profiles from AI usage, and we do not combine visitor behaviour across unrelated photographers to train models.

Inference workloads are pinned to European regions or vetted EU-based providers named in our subprocessor documentation. Logs are minimised to operate and debug the service. If regional availability changes, we will update subprocessors and provide notice consistent with our DPA.

AI suggestions are statistical and may be wrong, incomplete, or biased toward frequent compositions seen in training distributions. They are creative shortcuts, not statements of fact or substitutes for your editorial judgement. You should review outputs before client delivery, especially where sequencing affects emotional pacing or contractual deliverables.

We do not use your clients' personal data, favourites, or download histories to train general models. Events may be processed to render analytics you enable as controller and for security abuse detection. Where pseudonymisation is feasible for diagnostics, we apply it.

Prompt and output logs needed for reliability are retained only as long as necessary for those purposes, then deleted or aggregated. Security measures applicable to AI endpoints match the rest of the platform baseline described in our security documentation. If we discover misuse of AI endpoints, we may throttle or suspend offending accounts.

AI capabilities included in your plan today may evolve. If we introduce metered billing for heavy inference, we will provide at least thirty days' notice and preserve a commercially reasonable path to continue without paid AI where technically feasible. Promotional language on the website is illustrative only; entitlements are defined at checkout and in-product plan screens.

You can disable AI per gallery or account-wide in settings. Disabling stops new inference; previously saved edits remain unless you remove them. We periodically review default prompts and parameters to reduce unnecessary token usage and cost for you and environmental footprint for the community.

You may not attempt to extract underlying model weights, reverse engineer proprietary prompts at scale, use the service to generate unlawful content, or automate requests in a way that degrades stability for others. We may rate-limit abusive patterns.

AI outputs are provided "as is" to the maximum extent permitted by law without warranties of merchantability, fitness for a particular purpose, or non-infringement of third-party rights in suggestions. You should perform your own clearance for any text or sequencing you publish commercially. Nothing in these AI Terms limits liability for death, personal injury caused by negligence, fraud, or other liability that cannot be excluded by applicable law.

Provisions that by their nature should survive termination — including limitations on training use, confidentiality of logs, and dispute resolution references in the Terms — survive. If a competent court finds a clause unenforceable, the remainder stays in effect with the minimum modification necessary.