Skip to main content
Pre-release

This release is a pre-release and may not be stable for production use.

Immich Memories

CI codecov OpenSSF Scorecard Release Python License Docs

Watch your memories again, in films you can make your own.

Immich Memories turns the photos and videos in your Immich library into MP4 films with titles and music. Pick a month, a year, a trip, an album or a person. Review the proposed cut, swap what you don't like, then render. A quiet month makes a short film rather than 3 minutes of filler.

It runs on a plain NAS with no GPU and no model server. Basic reads the pictures with small local classifiers, GPU adds picture descriptions and extra checks, and Full adds a text reader that refines the cut.

Quick start · Documentation · Install guides · Demo

Choose a memory, review and change its cut, and watch the finished film
▶ Play the demo with music · Watch a finished trip film · Wikimedia Commons photographs under CC0 or CC BY, credited here · Documentation

Make your first film

Docker Compose is the recommended install. Start with Basic if you are unsure. Every tier makes a complete film, and you can change tiers later.

Tier What it adds What it needs Install
Basic Dates, favourites, people, places and small picture classifiers guide the cut; up to 1080p. 2 CPU cores and 4 GiB RAM for the app, in addition to Immich. Follow the steps below.
GPU Picture descriptions, document/sensitive-content checks and a family-viewing pre-screen. NVIDIA services with the Container Toolkit, or native Apple Silicon. Allow 8 GiB for the Compose app plus its model services. Select GPU in Quick start, or use native Mac.
Full A text reader refines the cut. The GPU setup plus an enabled reader with a 32k context window. Select Full in Quick start.

You need an existing Immich server (2.7.5 or later on v2, 3.2.4 or later on v3), Docker Compose v2 and disk space: about 25 GB of app data plus 0.35 GB per 1,000 pictures and videos in your library, plus the images and your finished films. The deployment matrix records what has been tested. GPU services can run on another machine. On a NAS container manager, use its platform guide.

  1. Download the two Basic files into a new folder:

    mkdir -p immich-memories && cd immich-memories
    VERSION=1.0.0-rc.10
    curl -fLO "https://github.com/sam-dumont/immich-memories/releases/download/v${VERSION}/docker-compose.yml"
    curl -fLO "https://github.com/sam-dumont/immich-memories/releases/download/v${VERSION}/example.env"
    cp example.env .env
    

    example.env pins the app image to the same release. Keep TIER=basic.

  2. Edit .env with your Immich address, API key and timezone:

    IMMICH_URL=http://192.168.1.10:2283
    IMMICH_API_KEY=your-api-key
    TZ=Etc/UTC
    

    In Immich, create the key under Account Settings → API Keys with the ten read permissions. Use an Immich address the container can reach; localhost means the container itself.

    To open the app from another computer on your LAN, also set this in .env:

    UI_BIND_ADDRESS=0.0.0.0
    

    The app has its own login, separate from Immich. If you set none, the first start generates one (step 4). To choose your own, add IMMICH_MEMORIES_AUTH_USERNAME and IMMICH_MEMORIES_AUTH_PASSWORD (at least 12 characters). LAN access uses HTTP; HTTPS setup is separate. Otherwise, keep the default localhost binding.

  3. Prepare the output folder and start the app. On Linux:

    mkdir -p output
    sudo chown 1000:1000 output
    chmod 600 .env
    docker compose up -d
    

    The app writes films as UID/GID 1000. For a Synology bind mount, use the DSM folder permissions instead of chown.

  4. Open the app at http://localhost:22830, or http://your-server-address:22830 for LAN access. Sign in as admin with the generated password, printed once in docker logs immich-memories (or use the login you set). On Memory, click Download models and wait for the card to disappear. No card means the files are already there.

    Make your first film from an album of 20–50 photos and short videos: choose Album, set the length to 0.5 minutes, then Cut, review and Render. Leave upload off and use Download film to keep the MP4. A full month can take hours to prepare on a NAS; a small album gets you a film first.

For a failed step, use Installation help. Once the first film works, After install covers home, people, language, backups and automation.

What you can do

  • Photos and videos in one film. Stills get animated, videos give their best few seconds, and a Live Photo plays as motion when its clip lasts at least 2.5 s and actually moves. Favourites and named people win close calls. Every shot says why it's in the cut, and the pool shows everything that was left out.
  • Change the cut before rendering. Remove a shot, trim a video, swap a picture or pull one in from the pool, then render that revision. "Never use" is a lasting decision, separate from dropping a shot from one film.
  • Titles and music out of the box. A film is titled by what it is and when, in the film's language: "Summer 2025", "Kit's birthday · 2026", "Kit & Pat · 2025", "Christmas · 2019–2025". Those are the app's own titles, never a model's. Model-written titles and summaries are an experimental opt-in (--llm-title). Every film gets a bundled track, or your own music folder. Maps and generated music are optional extras with their own setup.
  • Automate it later. A daily timer picks one memory worth making (a trip, a birthday, last month) and can upload it to Immich, with its title and story summary as the description, or keep it on disk. Tags for the type, years and stories are opt-in (upload.film_tags). Make a few films by hand first.
  • Ask in your own words. "our cat through the years" can become a film. Highly experimental: tuned on one real library, so preview before you trust it.
  • Add a GPU or a model when you want. Basic makes complete films on a CPU. A GPU adds captions and extra picture checks, Full adds a text reader on top. A hardware encoder speeds up rendering and changes nothing about which pictures make the cut.

How it chooses explains the selection rules. Improve a film covers length, missing people and shot changes. What doesn't work yet is in the issue tracker.

Access and privacy

The app talks to Immich with a scoped API key and never modifies your originals. Uploading a film is opt-in. There is no telemetry, and the update check only runs when you ask for it. Model downloads, optional model servers, maps, notifications and workers each have their own data flow, and a local URL alone doesn't prove nothing leaves. Privacy lists every recipient and how to check the Basic tier is offline.

Sign-in is on by default. The first start generates a password: find it in the container log (docker logs immich-memories). Lost it? immich-memories auth reset-password makes a new one. Native installs and the shipped Compose file listen on localhost only; change the port mapping and the sign-in is what stands between your LAN and your library. Operate and configure covers login, NAS/Kubernetes setup, backups and storage. Keep one UI replica. Report vulnerabilities as SECURITY.md describes.

This is an independent companion to Immich, not made or endorsed by the Immich team.

People and calendars

Families, relationships and calendars differ a lot, and this doesn't cover them all. If yours is missing, say so. Calendar limits and contributing household examples show where help is useful.

Development

Claude and Codex write most of the code; I set the direction and test films on a real library. make dev installs the development tools and make ci runs the checks. CONTRIBUTING.md describes the process; DISCLAIMER.md describes its limits.

License

MIT, see LICENSE.

Metadata

Release files for immich-memories 1.0.0rc11

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for immich-memories 1.0.0rc11
File Size Uploaded Provenance
immich_memories-1.0.0rc11.tar.gz 71.5 MB With provenance Details

Built distribution (wheel)

Table of built distributions (wheels) for immich-memories 1.0.0rc11
File Interpreter ABI Platform
immich_memories-1.0.0rc11-py3-none-any.whl Python 3 none any Details

Total release size: 82.4 MB

Release files / immich_memories-1.0.0rc11.tar.gz

Download URL immich_memories-1.0.0rc11.tar.gz
Size 71.5 MB
Tags Source
SHA-256 checksum
How to use checksums
21e8f5501fba5741998937df3e9674deac4771b6305b234965cfde4af0001cbf
BLAKE2b-256 checksum
How to use checksums
cc1a4effe1fa30c055a71659fa21a84fae477efeb7af6d71f5b1a715bff09afd
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Oct 24, 2026.

Transparency log

Release files / immich_memories-1.0.0rc11-py3-none-any.whl

Download URL immich_memories-1.0.0rc11-py3-none-any.whl
Size 10.9 MB
Tags Python 3
SHA-256 checksum
How to use checksums
051000b3e7bf0017351b009c70f89f30559b48cd6dec4a26800a55d8cf2945e6
BLAKE2b-256 checksum
How to use checksums
b699b4a1e3c8814059a449d42015c77888730e3f0466a83884bd32874894fc18
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Oct 24, 2026.

Transparency log

Security for immich-memories 1.0.0rc11

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Full provenance

All 2 files in this release have been uploaded with provenance metadata.

Select from 2 files

Attestation for immich_memories-1.0.0rc11.tar.gz

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Oct 24, 2026.

Transparency log

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Release history Release notifications | RSS feed

This release
Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page