When you upload a picture, it is written to your account's storage. When you start a workflow, our servers send the picture and your instructions to the AI system that runs that workflow. It produces the result; our servers copy it into your gallery; your browser reads it from there. The input is used for that generation and nothing else.
Generation runs on dedicated inference infrastructure, separate from the application servers and reached through private API calls. It receives what a generation needs, for the time a generation takes, and is bound by the same rule as everything else on this page: nothing it receives is kept, and nothing it receives is used for training.
We do not publish which models and systems run our workflows, nor how they are built, trained or operated. That changes as the field moves, and it is part of what we build. What does not change is the rule above: your content is never training material, here or anywhere along the way.
Visual DNA is the one place where your images shape a model: it learns your brand's style from the pictures you give it, for your account only. It is never shared with another account and it is deleted with your account.