Damselfly

Last reviewed Sep 13, 2026 · 831 words

Damselfly writes your tags into the photos. When you keyword an image, it uses ExifTool to store that keyword in the file's IPTC and XMP metadata, so Lightroom, digiKam and every future tool you ever use will see it. Immich and PhotoPrism keep tags in their own database, and if you leave them the tags stay behind. That single design decision tells you who Damselfly is for: photographers with a large, folder-organised archive who want fast search and durable tagging, and who do not want a photo app to become the owner of their pictures.

An index over your folders, not a library that swallows them

Damselfly is GPL-3.0, written in C#, first released in 2020, with 1,787 GitHub stars at the time of writing. You point it at an existing directory tree and it reads it, read-only if you mount it that way, and builds its own index in a separate config directory. Nothing is moved, renamed or converted. Delete the container and the archive is exactly as you left it, plus any keywords you asked it to write.

That makes it the right shape for a specific kind of user. If your photos already live in 2019/2019-06-Portugal/ style folders, exported from a camera workflow, with RAW and JPEG side by side, Damselfly respects that and searches across it. If your photos are a phone camera roll you want automatically backed up and shared with family, that is Immich, and Damselfly will feel like the wrong tool within a day. The two are not competing for the same person.

Search speed on 500,000 images is the headline

The project's stated design goal is fast search across very large collections, and it delivers: the index is built for quick keyword, date, folder and metadata queries over hundreds of thousands of files without the interface bogging down. Face recognition and object detection run on the server using bundled models and feed the same search, so "photos of Anna in 2021 with a dog" is a query rather than a project. Results can be selected in bulk and exported, downloaded as a zip, or handed to a basket for tagging, which is the workflow: search, select, tag, and the tags land in the files.

Compared with PhotoPrism, which also indexes existing folders, Damselfly is narrower and faster at the archive-management job and weaker at being a pleasant gallery to browse on a Sunday. The Immich versus PhotoPrism comparison covers the two bigger projects; Damselfly is the specialist beside them.

The deployment is one container and two mounts

services:
  damselfly:
    image: webreaper/damselfly:latest
    ports:
      - "6363:6363"
    volumes:
      - /mnt/photos:/pictures:ro
      - ./damselfly-config:/config
    restart: unless-stopped

Port 6363 is the default. Mount /pictures read-only until you decide you want tag writing, then remount read-write; that is the one moment Damselfly touches your originals. The /config volume holds the SQLite index and thumbnails, and the thumbnail cache grows with the archive, so put it on a disk with room. 1 GB of RAM is the honest floor for a large library while the first index runs; face recognition and object detection are CPU work and the initial pass over 200,000 images takes hours, not minutes. A quad-core machine finishes it overnight.

There is also a desktop client that syncs thumbnails and lets you tag from a laptop. Most people never install it, and the web interface is enough.

The niche cuts both ways

A tool built for one job by a small team has one maintainer's priorities and a small user base to find bugs. Releases come in bursts. The mobile experience is a responsive web page, not an app, and there is no automatic upload, no sharing links for relatives and no multi-user model to speak of; the catalogue tags it single-user, and that is accurate. If any of those matter, Immich is the safer pick and the photos category will show you why it dominates.

What the niche buys is that Damselfly does nothing to your archive you did not ask for, and every keyword you add outlives the software. For someone who has already been burned by a photo app that held their tags hostage, that is the whole argument.

What I'd do

If I were a photographer with a 300,000 image folder archive, I would run Damselfly against it read-only for a month, let the recognition finish, and see whether the search earns its keep. If it did, I would flip the mount to read-write, write keywords into the files, and keep digiKam or Lightroom for editing. I would not use it as the family's phone backup; that job goes to Immich on the same server, pointed at a different folder. Two tools, each owning nothing, is the setup that survives the next decade.