v3 Video Transcode Jobs silently failing

[okh-mzny]

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

• Yes

The bug

Since the update to Immich v3, transcode jobs do not run properly anymore.
Attempting to queue up all videos for processing briefly shows the job counter go to the expected number, then immediately fall to 0.

This seemingly happens regardless of settings.
I have tried and verified the following, without any luck:

• Tried changing transcode policy, even on "all" no jobs are completed
• Tried disabling HW acceleration, both entirely and only for decoding
  • (Usually transcoding is running on an Intel Arc A380)
• Tried removing the encoded video folders
  • (Broke playback for existing transcoded videos, and no transcodes were ever (re)created)
• Verified my GPU is working and detected
  • (ML jobs are unaffected and are running fine on the GPU)
• Verified FFmpeg is available and able to use the GPU inside the container
  • (test encode is working and encode activity is visible in nvtop)
• Tried the new Real-Time transcoding
  • (Error 404 when trying to get the .m3u8 playlist, playback broken)

The OS that Immich Server is running on

Docker (OpenSUSE MicroOS Host)

Version of Immich Server

v3.0.1

Version of Immich Mobile App

N/A

Platform with the issue

• Server
• Web
• Mobile

Device make and model

No response

Your docker-compose.yml content

version: '3.8'

services:
  immich:
    image: ghcr.io/imagegenius/immich:openvino
    container_name: immich
    env_file:
      - stack.env
    environment:
      - PUID=1000
      - PGID=1000
      - TZ=Etc/UTC
      - DB_HOSTNAME=postgres
      - DB_USERNAME=postgres
      - DB_DATABASE_NAME=immich
      - REDIS_HOSTNAME=valkey
      - DB_PORT=5432
      - REDIS_PORT=6379
      - SERVER_HOST=0.0.0.0
      - SERVER_PORT=2883
      - MACHINE_LEARNING_MODEL_TTL=0
      - IMMICH_LOG_LEVEL=verbose
    volumes:
      - immich-config:/config
      - /srv/immich/photos:/photos
    ports:
      - 2883:2883
    device_cgroup_rules:
      - 'c 189:* rmw'
    devices:
      - /dev/dri:/dev/dri
    restart: unless-stopped

  valkey:
    image: docker.io/valkey/valkey:9
    container_name: valkey
    restart: unless-stopped 

  postgres14:
    image: ghcr.io/immich-app/postgres:14-vectorchord0.4.3-pgvectors0.2.0
    container_name: postgres
    env_file:
      - stack.env
    environment:
      POSTGRES_USER: postgres
      POSTGRES_PASSWORD: ${DB_PASSWORD}
      POSTGRES_DB: immich
    volumes:
      - postgres-data:/var/lib/postgresql/data
    restart: unless-stopped 

volumes:
  immich-config:
  postgres-data:

Your .env content

DB_PASSWORD=secret

Reproduction steps

1. Attempt to start any video transcoding job on v3
2. It fails silently and no actual transcode ever occurs

Relevant log output

--- Hitting "Transcode All" ---

[Nest] 272  - 07/04/2026, 1:33:37 PM   DEBUG [Api:QueueService~uf98nr8d] Handling command: queue=videoConversion,command=start,force=true
command=start command=start force=true
[Nest] 272  - 07/04/2026, 1:33:37 PM   DEBUG [Api:LoggingInterceptor~uf98nr8d] PUT /api/jobs/videoConversion 200 1.44ms
[Nest] 272  - 07/04/2026, 1:33:37 PM VERBOSE [Api:LoggingInterceptor~uf98nr8d] 

--- No further log related to transcode, yes the last line of log is printed empty ---

Additional information

I'm unsure how to approach this further, as there is no error logged even with verbose logging. If someone could help me debug this it would be much appreciated.

[Mraedis]

What is your host hardware? Also, imagegenius is not an official install.

[okh-mzny]

The CPU type is set to host passthrough.

[Mraedis]

Seems good to me, could you check out the imagegenius git and see if the issue is reported there?

[arthyk]

Hi everyone,
​I am experiencing the exact same behavior on a fresh setup running Immich v3+ (consolidated immich-server architecture), and I would like to provide detailed logs and architecture context to help debug this issue.

​Environment & Hardware:
​OS: Ubuntu Server / Debian based Linux
​Deployment: Docker Compose (v3 consolidated template)
​CPU: Intel Core i5 (11th Gen) with Intel Iris Xe Graphics
​Hardware Acceleration: Intel QuickSync configured via hwaccel.transcoding.yml (Using /dev/dri/renderD128, GID 993). Docker execution confirms containers have proper card0, card1, and renderD128 access with read/write permissions.
​Storage Architecture (Hybrid Setup):
​Main Library (/data/library): Mounted via NFS from a remote NAS (x-systemd.automount active).
​Cache & Transcoding Directories (/data/encoded-video, /data/thumbs, etc.): Local High-Speed NVMe SSD (/home/homelab/immich-app/immich-ssd-cache/).
Image: ghcr.io/immich-app/immich-server:${IMMICH_VERSION:-release}

​The Symptoms & Logs:
​The Transcoding Fail / ENOENT Error:
When attempting to trigger video transcoding (via Jobs -> Transcoding -> Missing) or when playing an asset that requires a transcoded version, the engine completely skips or fails to generate the .mp4 file in the local SSD cache.

When I open a video in app or browser. I get this error ENOENT in the immich-server:

[Nest] 25 - ERROR [Api:LoggingRepository] Unable to send file: Error: ENOENT: no such file or directory, access '/data/encoded-video/f4128cf2-152f-483e-805b-fe5255a62743/b1/15/b11534d5-c909-4bff-8819-aeb1b78ea3b7.mp4'
Error: ENOENT: no such file or directory, access '/data/encoded-video/f4128cf2-152f-483e-805b-fe5255a62743/b1/15/b11534d5-c909-4bff-8819-aeb1b78ea3b7.mp4'
at async access (node:internal/fs/promises:607:10)
at async sendFile (/usr/src/app/server/dist/utils/file.js:41:9)
at async AssetMediaController.playAssetVideo (/usr/src/app/server/dist/controllers/asset-media.controller.js:82:9)

Note: Folder permissions on the host for /immich-ssd-cache have been forcefully set to 777 with chown 1000:1000 to completely rule out local permission blocks.

[okh-mzny]

I'm seeing an interesting parallel, while I don't have qutie the same hybrid storage setup like you do, I do also use a network share (samba) to store my library.

I don't differenciate between thumbs, encoded video and assets, I only have the database on local SSD storage (VM) while the rest lives on the large samba share.

I didn't bring it up earlier because I did not think it would be important (a file is a file afterall, and my existing transcodes do work)

[akramer]

I also have my library stored on a samba share, and I'm getting a 404 when trying to real-time transcode (I don't do offline transcodes)

[Nest] 37  - 07/07/2026, 3:21:25 AM   DEBUG [Api:LoggingInterceptor~g7bztt5c] GET /api/assets/7787df11-9561-42bd-9cdc-
9da4c62557fb/video/stream/main.m3u8 200 2.70ms 192.168.0.1
[Nest] 37  - 07/07/2026, 3:21:25 AM   DEBUG [Api:GlobalExceptionFilter~g7bztt5c] HttpException(404): {"message":"Asset metadata is not yet ready for streaming","error":"Not Found","statusCode":404}

[okh-mzny]

I found a solution for this, rerun metadata extraction for all assets, and then transcoding will just work again.

The issue, from my limited understanding, seems to come down to v3 relying on video metadata being present to kick off transcoding jobs, instead of previously directly probing the file and then deciding whether to transcode it.
In my case, I'm assuming that metadata was missing for whatever reason and had to be created by metadata extraction

The library being on a network share is entirely unrelated.

Being short on time and clueless on how to debug this without proper logs, I handed the job off to an LLM to find possible regressions between v2 and v3 for me.

For what its worth, here is the AI analysis:

⚠️ AI generated, may be wrong

Relevant versions: v2.7.5 -> v3.0.0 -> v3.0.1

Summary
- I did find a major v3 architectural change that can cause transcode jobs to drain immediately with little/no logging if new video metadata rows are missing.

Key v3 code changes
1) Transcode job no longer probes source file directly at transcode time
- v2.7.5: server/src/services/media.service.ts
  - handleVideoConversion() called mediaRepository.probe(input, ...)
  - chose main streams during the transcode job itself
- v3.0.0+: server/src/services/media.service.ts
  - handleVideoConversion() now uses asset.videoStream, asset.audioStream, asset.format from DB-loaded metadata
  - if metadata is missing after a successful asset load, it logs:
    "Skipped transcoding for asset ${asset.id}: missing metadata; re-run extraction first"
  - important nuance: this log only happens if getForVideoConversion(id) returned an asset object in the first place

2) getForVideoConversion() became metadata-dependent in v3
- v2.7.5: server/src/repositories/asset-job.repository.ts
  - getForVideoConversion(id) selected asset + files only
- v3.0.0+: same file
  - getForVideoConversion(id) now does:
    - innerJoin('asset_exif', ...)
    - innerJoin('asset_video', ...)
    - leftJoin('asset_audio', ...)
  - If asset_exif or asset_video row is absent, query returns no row.

3) Silent failure path in v3 transcode jobs
- server/src/services/media.service.ts
  - handleVideoConversion():
    const asset = await this.assetJobRepository.getForVideoConversion(id);
    if (!asset) {
      return JobStatus.Failed;
    }
- This branch has no log line.
- Therefore symptom can be:
  - queue count rises
  - jobs are consumed
  - queue count drops to 0
  - no transcoding starts
  - no "missing metadata" message appears
- This also explains why the warning
  "Skipped transcoding for asset ${asset.id}: missing metadata; re-run extraction first"
  may never appear: the job can fail earlier because getForVideoConversion() returned no row at all.

[Mraedis]

Interesting, thanks for probing. @arthyk can you try queue'ing metadata extraction first as well?

[CarlvO]

Had the same issue with some videos and forcing Metadata extraction fixed it for me as well.

Thanks @okh-mzny

[CarlvO]

Reflecting on this a bit, I believe we should request an enhancement in Immich. First of all, I have no idea why meta data weren´t present for some of my videos. So that leads me to think more users may face this.

Graceful behaviour by the application IMO would be that the user gets a message when they order a re-transcode of a video if no meta data are found for it so they are aware that meta data are missing. If possible, it would be better to first extract the meta data of the asset and then proceed to perform the transcode.