Upload and classification
- Drop zone for new photos
- Manual confirmation of theme, collection and quality
- Reject / approve flow before photos enter the catalogue
- Notes that remain with the photo
Case study / AI photo dashboard
Smart AI Curator
A photo archive dashboard that combines editorial work, AI classification and embedding search. Instead of yet another folder system, the photographer gets a tool for finding series, building collections and preparing a public portfolio.
- Product
- Smart AI Curator
- Type
- AI dashboard for photographic archive
- Scope
- Dashboard, API, AI pipeline
- Year
- 2026
The problem is not the upload. The problem is curation.
Photo archives quickly grow too large to sort by hand. Folders remember dates and file names, but not atmosphere, visual kinship, the potential for a series or a photo's place in a future exhibition.
So Smart AI Curator is built as a photographer's workbench: a new photo comes in through the inbox, gets analysed, goes into the library, links up with similar work and then moves on to collections, the timeline or the public preview.
What the dashboard does.
Curated series
- Generating collections from embedding space
- Custom collections from text or seed photos
- Snapshot history and restore flow
- Narratives for exhibition sets
Visual map of the archive
- Vertex AI multimodal embeddings
- 3D PCA view of the library
- Colour by theme, collection or quality
- Orbit control and cluster overview
Find related shots
- Cosine similarity over embeddings
- Suggested photos around the seed selection
- Threshold control for tighter or looser series
- Iterative collection expansion
Archive through time
- Monthly work review
- Quick access to photos by period
- The link between date, theme and collection
- Control over catalogue changes
Portfolio before publishing
- A preview of how the collection looks in public
- Philosophy and description for each series
- Masonry and grid view
- A bridge between the dashboard and the portfolio website
AI is an assistant, not a replacement for the editor.
The system uses embeddings to turn the archive into a searchable space, but the final decision stays with a person. The photographer sets thresholds, approves suggestions, edits collections by hand and can restore an earlier version of the structure if the AI's direction is off.
That's a key product decision: the dashboard doesn't pretend to be an automated CMS. It speeds up the slowest part of the work, reveals connections that are hard to see in folders and leaves enough control for the result to stay the photographer's own.
What it does under the hood.
Every photo passes through a Vertex AI multimodal model that turns it into a 1408-dimensional vector. That vector is the basis for everything else: similarity search, clusters, the 3D map and automatic collection generation.
To browse the archive, PCA projection reduces the high-dimensional space to three axes you can rotate and zoom. Collections are built by cosine similarity around a seed selection, and every version of the structure is saved as a snapshot that can be restored. The result is a searchable space where the editor works fast and nothing gets lost.
A new photo lands in the inbox
Uploads don't go straight into the public catalogue. They first go through review, classification and quality selection.
Embeddings map visual relationships
Each photo gets a vector that enables similarity search, clusters and a 3D view of the whole archive.
AI suggests a collection
The system can generate series, expand a seed selection and write a narrative, but the user confirms the result.
Preview prepares the public portfolio
A curated series is previewed in a format close to the public website, before publishing or sending it to a client.
What the dashboard does
Twelve modules, one flow.
12working modules
1408dim / embedding vector
3Darchive map
6curation phase
01
Inbox dropzone
Import new photos with manual confirmation of theme, collection and quality.
02
Approve / reject flow
Selection before photos enter the catalogue, with notes that stay with each photo.
03
Library
Central catalogue with ratings, metadata and quick filtering.
04
Multimodal embeddings
A 1408-dimensional Vertex AI vector for every photo in the archive.
05
3D archive map
PCA projection of the library with orbit control and colouring by theme.
06
Similarity search
Cosine similarity with a threshold control for tighter or looser series.
07
AI Collections
Generating series from the embedding space or from text and seed photos.
08
Snapshot and restore
Collection history and rollback to an earlier structure without losing work.
09
Timeline
Monthly overview of the archive and quick access to photos by period.
10
Loupe view
Detailed photo view with metadata and EXIF data.
11
Portfolio preview
Masonry and grid view of a collection before it goes public.
12
Series narratives
Philosophy and description for each collection, ready to publish.
