Facial recognition is not part of Pencil. The platform runs no facial recognition technology when you upload, generate or edit images, and it does not identify or match the people who appear in your assets.
How does Pencil handle images that contain people?
Generative edits such as resizing, reformatting and background changes apply to the whole image by default, regardless of its content. Pencil does not detect people in an image and treat them differently from any other element, so an edit is not guaranteed to leave a person untouched. Always review outputs where people appear.
How can I keep people unchanged during an edit?
Prompting is the only way to steer an edit away from a person. For example: "change only the background, do not change the person". This works precisely in most cases. Frontier models like GPT Image 2 and Nano Banana Pro natively understand what an image contains and usually handle this sort of task well. But users need to understand that, due to the non-deterministic nature of generative AI, it's always a steering instruction, and does not offer a watertight guarantee.
Agents and workflows can add a further safeguards. If your team regularly works with photography of people, you can set up an agent to check whether an image contains people before a generation step runs, and handle those images according to your rules.
Different companies have different rules around the use of generative AI. A tech company, for example might have no problem with AI close-ups of human faces in their ads. On the other hand, a makeup company might insist on real humans for any close ups featured in ads, yet allow AI-generated depictions of people in at a distance in imagined settings. Some companies prefer to avoid any AI depictions of human figures altogether and use AI only for landscape, environment or conceptual shots.
It's up to Pencil users to make sure they are using generative AI responsibly, and producing content according to their own brand and company's guidelines.