A personal archive is almost never organized like a library. It is made of photographs from different phones, untitled clips, tests, scans, voice messages, screenshots and folders created for projects that were later abandoned. For years, the problem was simply finding an image again. Now visual-search tools and generative models promise something more: interrogating this material, recognizing recurrences, bringing distant fragments together and suggesting possible narrative paths.
That promise needs to be read precisely. AI does not possess a person’s memory and does not know why a photograph matters. But it can describe what it sees, search for people, objects, places or similar atmospheres, group images and make large quantities of otherwise invisible material explorable. Google Photos and Apple Photos already show this logic in their search functions: the process does not always begin with the file name, but with what the image contains. For anyone who works with images, this simple possibility changes the first gesture of research.
The creative leap comes when search is not treated as an automatic catalog but as a phase of editing. A photographer can ask to gather every image containing a particular kind of light, a corridor, a profile or a recurring object. A director can surface, across years of footage, moments of waiting, returns and everyday gestures they had not noticed. The system offers associations; the author decides whether they are coincidences, clues or simply mistakes. The archive begins to produce a story only when someone chooses a rhythm and a distance.
This distinction also protects against the rhetoric of automatic discovery. A model can find images that are formally similar but belong to completely different moments in a life; it can confuse a person, misread a place or assign intention to a detail. Visual similarity is not yet meaning. That is why AI is more useful as a lens for exploration than as an autonomous narrator: it expands the field but does not decide what a relationship should mean. The value of an archive also lies in its gaps, uncertain dates and images that resist perfect classification.
Working with an archive also requires attention to privacy. Personal photographs contain faces, places, habits and people who did not choose to enter an analytical system. Before uploading material to a platform, one needs to know where it will be stored, who can access it, whether it may be used for training and how it can be removed. Public projects also require releases, consent and risk assessment for the people appearing in the images. Making an archive easier to access should not automatically make it more exposed.
There is also the issue of provenance. When historic photographs, personal images and new AI generations coexist in the same film or installation, the audience should be able to understand what it is seeing. Standards such as C2PA and Content Credentials offer a way to associate information about origin and modifications with a file. They do not solve every problem of authenticity on their own, but they help build a grammar of credits: document, reworking, reconstruction, generated image. Clarity does not reduce the poetry of the material; it allows each step to be appreciated more accurately.
Perhaps the most interesting possibility is that AI can restore dignity to minor material. Not only the beautiful image or exceptional moment, but the out-of-focus shot, the passage between two rooms, light from a window, a voice recorded by accident. In a traditional archive, these fragments are often forgotten because they have no keyword. If visual search brings them back to the surface, editing can use them to build continuity, atmosphere or contradiction. It is not a machine inventing a better memory; it is a system making the traces from which memory can be built available again.
ACTIVE MEMORY is therefore an invitation to think of the archive as a workspace rather than a storage room. Technologies can organize, search and propose; they cannot replace the responsibility of choosing what to preserve, what to show and what to leave silent. The possible film is not already hidden inside the data. It is born from the encounter between recovered material, human care and the decision to give those images a new form.