Facial recognition not working

[HipyCas]

I have searched the existing issues, both open and closed, to make sure this is not a duplicate report.

• Yes

The bug

Since beginning with Immich (quite a long time ago) everything has worked without flaw, including facial recognition. Suddenly, I realized that since around January 22nd 2025, facial recognition stopped working. That didn't prevent manually adding faces when the feature was introduced, but for some reason it does not automatically recognize them neither for new nor old assets.

After looking into the logs, I thought that the problem might be with the upload & the metadata from the phone & tablet, but there are some photos uploaded through the web manually that did not get facial recognition either way.

The logs pasted look to me like they may be related, but I am not 100% sure of that. They are from the server container, the machine learning container does not output any logs other than the startup ones (which look like the server restarted this night, which may make sense as we have Watchtower configured for automatic updates).

(Even thought we have automatic updates with Watchtower, every morning I manually check the releases page if there has been an update to fix any possible breaking changes or similar)

The OS that Immich Server is running on

Debian

Version of Immich Server

v1.131.3

Version of Immich Mobile App

v1.131.1

Platform with the issue

• Server
• Web
• Mobile

Your docker-compose.yml content

immich-server:
    image: ghcr.io/immich-app/immich-server:release
    container_name: immich-server
    volumes:
      - ${UPLOAD_LOCATION}:/usr/src/app/upload
      - /etc/localtime:/etc/localtime:ro
    environment:
      - NODE_ENV=production
      - DB_USERNAME=${DB_USERNAME:?immich server db username not set}
      - DB_PASSWORD=${DB_PASSWORD:?immich server db password not set}
      - DB_DATABASE_NAME=${DB_DATABASE_NAME:?server db name not set}
      - JWT_SECRET=${JWT_SECRET}
      - ENABLE_MAPBOX=${ENABLE_MAPBOX}
      - MAPBOX_KEY=${MAPBOX_KEY}
      - IMMICH_TELEMETRY_INCLUDE=all
    depends_on:
      - immich-redis
      - immich-database
    env_file:
      - .env
      - .immich.env
    ports:
      - ____:2283
      - ____:8081
      - ____:8082
    restart: unless-stopped
    labels:
      - "com.centurylinklabs.watchtower.enable=true"

  immich-machine-learning:
    image: ghcr.io/immich-app/immich-machine-learning:release
    volumes:
      - ${UPLOAD_LOCATION}:/usr/src/app/upload
      - /____:/cache
    environment:
      - NODE_ENV=production
    env_file:
      - .env
      - .immich.env
    expose:
      - 3003
    restart: unless-stopped

  immich-redis:
    container_name: immich_redis
    image: redis:6.2
    healthcheck:
      test: redis-cli ping || exit 1
    restart: unless-stopped
    mem_limit: 2G
    labels:
      - "com.centurylinklabs.watchtower.enable=true"

  immich-database:
    container_name: immich_postgres
    image: registry.hub.docker.com/tensorchord/pgvecto-rs:pg14-v0.2.0@sha256:90724186f0a3517cf6914295b5ab410db9ce23190a2d9d0b9dd6463e3fa298f0
    environment:
      POSTGRES_PASSWORD: ${DB_PASSWORD}
      POSTGRES_USER: ${DB_USERNAME}
      POSTGRES_DB: ${DB_DATABASE_NAME}
      PG_DATA: /var/lib/postgresql/data
      POSTGRES_INITDB_ARGS: '--data-checksums'
    volumes:
      - ${DB_DATA_LOCATION}:/var/lib/postgresql/data
    healthcheck:
      test: pg_isready --dbname='${DB_DATABASE_NAME}' --username='${DB_USERNAME}' || exit 1; Chksum="$$(psql --dbname='${DB_DATABASE_NAME}' --username='${DB_USERNAME}' --tuples-only --no-align --command='SELECT COALESCE(SUM(checksum_failures), 0) FROM pg_stat_database')"; echo "checksum failure count is $$Chksum"; [ "$$Chksum" = '0' ] || exit 1
      interval: 5m
      start_interval: 30s
      start_period: 5m
    restart: unless-stopped
    command: ["postgres", "-c" ,"shared_preload_libraries=vectors.so", "-c", 'search_path="$$user", public, vectors', "-c", "logging_collector=on", "-c", "max_wal_size=2GB", "-c", "shared_buffers=512MB", "-c", "wal_compression=on"]
    labels:
      - "com.centurylinklabs.watchtower.enable=true"

Your .env content

DB_HOSTNAME=immich_postgres
DB_USERNAME=____
DB_PASSWORD=____
DB_DATABASE_NAME=____

REDIS_HOSTNAME=immich_redis

UPLOAD_LOCATION=____

LOG_LEVEL=verbose

JWT_SECRET=____

ENABLE_MAPBOX=true
MAPBOX_KEY=____

TYPESENSE_API_KEY=____
# TYPESENSE_ENABLED=true

DISABLE_REVERSE_GEOCODING=false
REVERSE_GEOCODING_PRECISION=3

# IMMICH_WEB_URL=http://immich-web:3000
# IMMICH_SERVER_URL=http://immich-server:3001
# IMMICH_MACHINE_LEARNING_URL=http://immich-machine-learning:3003

Reproduction steps

There is no specific reproduction steps sadly, simply happened after updating.

Relevant log output

Error: Asset 1da55e59-0a0a-4c9a-8f6c-2b240d7e30bf dimensions are unknown
immich-server  |     at PersonService.getInputDimensions (/usr/src/app/dist/services/person.service.js:520:19)
immich-server  |     at PersonService.handleGeneratePersonThumbnail (/usr/src/app/dist/services/person.service.js:445:57)
immich-server  |     at process.processTicksAndRejections (node:internal/process/task_queues:105:5)
immich-server  | [Nest] 7  - 04/20/2025, 12:00:35 AM   ERROR [Microservices:{"id":"0bfbf32d-37f3-4908-8933-ecd94fe35b1a"}] Unable to run job handler (thumbnailGeneration/generate-person-thumbnail): Error: Asset 369f39d8-00e1-44fc-9576-cc00f32c9f4f dimensions are unknown

Additional information

No response

[bo0tzz]

I think this is the same issue as #16372?

[HipyCas]

The thing is, I don't have any problem with that (the people preview faces), but with the automatic tagging of them. As I said, maybe the log provided is not related to the problem, but it has been raised even though there has been no "updates" regarding people's thumbnail. Only thing I can think about it related is that maybe the pictures are properly tagged with the faces, but the web viewer does not show them as the thumbnail gives an error (some people are shown though on older photos). Is there any way (maybe CLI) through which I could maybe check if they are correctly checked?

I have also checked rn and manually changing the featured photo of a couple people works basically without flaw (it shows an "error loading image" in the image in the moment of change, but going back to the people view or clicking on it properly shows it, and no error log shows up).

[bo0tzz]

tbc the comments there show the same error in the logs as you're seeing.

[HipyCas]

That is true, but I'm not having any problems with the thumbnail generation, but rather what I mentioned of no facial recognition on new assets at all

[maelp]

I have similar issues, it worked once, the jobs seem to run, but then there's no face displaying, and when I click "missed" to run the facial detection / facial recognition on "new stuff" (even though I haven't added some since last time), it tries to run again on thousands of photos

[maelp]

(it used to work when I uploaded images using immich-cli, but now I'm using an "external library" that I'm mounting and it does the scanning but then there's no "people" being created)

[maelp]

In my case I have this in the logs

could it be due to a weird filename or something?

[maelp]

Hmmm might have been due to an out-of-date immich-machine-learning container image

[jtut1731]

I am a new user and just installed immich on unraid using docker compose method. I installed version 1.134.0, and did not set up any HW acceleration. Just left it to CPU. After uploading images with people, absolutely no faces were detected under the People view. Logs for machine_learning do not show any errors other than "Worker (pid:9) was sent SIGINT!".

[maelp]

@jtut1731 not sure if that's your issue, but for me the version of "immich" and "immich-machine-learning" images on my docker-compose were different, and that was the issue

[jtut1731]

I used the docker-compose.yml file from the release, and the images are defined as:

image: ghcr.io/immich-app/immich-server:${IMMICH_VERSION:-release}
image: ghcr.io/immich-app/immich-machine-learning:${IMMICH_VERSION:-release}

and from the example.env that I used:
IMMICH_VERSION=release

doing a docker image ls:

docker image ls
REPOSITORY                                   TAG                                  IMAGE ID       CREATED         SIZE
ghcr.io/immich-app/immich-server             release                              cc8195230205   21 hours ago    2.13GB
ghcr.io/immich-app/postgres                  14-vectorchord0.3.0-pgvectors0.2.0   d824c6d049d7   24 hours ago    741MB
ghcr.io/immich-app/immich-machine-learning   release                              87e5e4f325cb   2 weeks ago     1.04GB

I tried pinning the version like it says in the environment variables file. Same result. It does seem that both images were tagged with "release" as well as "v1.134.0", so I would not have expected any differences.

im having a similar problem, it was working perfectly but now when I try redo / rerun the features it doesnt work, I have reset everything back to the start to see if I could fix it that way as well as trying different AI models.

as you can see it runs and once done, nothing changes no faces are detected nothing under people

[jtut1731]

I did additional testing, and after uploading some selfies, I was able to detect and create a person. The Person configuration is set for a minimum of 3, and some of the already-uploaded photos did not include the whole face. So maybe a combination of it not detecting a face in some photos, and not having 3 faces of the same person may have led me to believe it wasn't working at all. It would help to improve the UI on a photo review so that you can run facial detection right then (not just re-queue it), and it indicate the faces it detects.