GitHub - ThatXliner/Hologram: Pro-grade photo gallery

5 min read Original article ↗

Hologram — Cull the shoot. Keep your files.

A local-first photo culling and management workspace for photographers who want control of their files.

RAW+JPEG pairing · keyboard-first culling · deep EXIF search · explainable local AutoCull

Build status Download nightly build Local-only processing

Website · Download nightly · Roadmap

Hologram reads the folders you already own and gives you a fast, dense workspace for turning a shoot into a set of keepers. It does not require a cloud library, proprietary photo vault, or heavyweight editing catalog.

Point it at a folder. Hologram streams previews, pairs RAWs with their camera JPEGs, exposes the metadata behind every frame, and keeps the next decision one key away.

Important

Hologram is under active development. Nightly builds are usable, but expect rough edges and evolving workflows.

See the whole shoot

Hologram library showing a dense photo grid, culling state, filters, and session progress

One uninterrupted workflow

Stage What Hologram does
Import Scans a normal folder and streams thumbnails progressively. Embedded RAW previews are used when available.
Browse Shows a dense adjustable grid or chronological timeline with filename, camera, exposure, pairing, rating, and flag state.
Cull Pick, reject, unmark, and rate without leaving the keyboard. Optional auto-advance turns review into one decision per frame.
Inspect Zoom from fit to 1200%, switch between RAW and JPEG, compare neighboring frames, and inspect full EXIF.
Organize Search filenames and metadata, compose filters, save searches, tag frames, and build opt-in local Smart Collections.
Deliver Export the visible set to a folder, ZIP, or Lightroom-ready structure with pair handling, renaming, and XMP sidecars.

Built for real camera files

  • Broad RAW support, including CR2, CR3, NEF, ARW, DNG, RAF, ORF, RW2, and more
  • RAW+JPEG pairs represented as one logical photograph
  • Embedded JPEG extraction for fast first previews
  • Progressive loading from thumbnail → embedded preview → full resolution
  • Local metadata and preview caches for repeat visits
  • Non-destructive preview adjustments and explicit export

AutoCull, without the oracle act

Hologram AutoCull showing similar-image clusters, recommendations, confidence, and score details

AutoCull groups bursts and visually similar frames, produces SELECT / MAYBE / REJECT / NEEDS_REVIEW recommendations, and exposes the signals behind each result. Local DINOv2 embeddings are combined with perceptual and technical features, then adjusted by an on-device preference model trained from your picks, ratings, and burst winners.

  • Review one cluster at a time or rank the entire shoot worst-first
  • Inspect technical, personal, final, and confidence scores independently
  • Adjust the reject threshold before applying any bulk action
  • Train and switch between named taste profiles
  • Keep manual flags authoritative over every recommendation
  • Run indexing and inference locally

Loupe, compare, adjust

Loupe and EXIF Preview-only adjustments
Hologram loupe view with full EXIF and culling controls Hologram non-destructive adjustment workspace

More views: RAW+JPEG pairing · filtered library · first run

The mouse is optional

The culling vocabulary stays available wherever a photograph is active.

Key Action
← ↑ → ↓ Previous / next photograph
P Pick
X Reject
U Clear flag
0–5 Set star rating
R Toggle RAW / JPEG
D Toggle comparison
E Toggle adjustments
F Fit photograph
+ / − Zoom in / out
O Open in the system viewer
Esc Close the current surface or reset zoom

Architecture

flowchart LR
    A["Folders on disk"] --> B["Rust + Tauri core"]
    B --> C["RAW preview pipeline"]
    B --> D[("Local metadata and preview cache")]
    C --> E["SvelteKit workspace"]
    D --> E
    E --> F["Local visual index + AutoCull"]
    F --> D
    E --> G["Folder / ZIP / Lightroom export"]
    E --> H["XMP sidecar round-trip"]
Loading
Layer Technology
Desktop shell and system integration Tauri 2
File scanning, RAW previews, metadata, caching, export Rust
Product interface SvelteKit 5 + TypeScript
Styling Tailwind CSS 4 + IBM Plex Sans / Mono
Local visual features Transformers.js + DINOv2 embeddings
Packaging and builds Bun + Vite + GitHub Actions

The boundary is intentional: filesystem-heavy and performance-sensitive work lives behind typed Tauri commands; the Svelte layer owns the dense interactive workflow.

Try it

The fastest path is the latest artifact from the main branch:

Download a nightly build

Nightly artifacts are produced for macOS, Windows, and Linux. They are development builds rather than polished releases.

Build from source

You will need Bun, a Rust toolchain, and the platform dependencies required by Tauri.

git clone https://github.com/ThatXliner/Hologram.git
cd Hologram
bun install
bun run tauri dev

Useful project commands:

bun run check        # Svelte and TypeScript diagnostics
bun run build        # Production frontend build
bun run screenshots  # Regenerate deterministic product screenshots

Principles

  1. Your files remain files. Hologram works with the filesystem instead of hiding a library inside a proprietary vault.
  2. Local means local. Photo analysis, visual indexing, and preference learning run on your machine.
  3. Automation must explain itself. Recommendations expose their inputs and always yield to manual decisions.
  4. Keyboard flow is product functionality.
  5. Photos are the hero. Density is useful; chrome should recede.

Status and roadmap

Hologram is being built in the open. The current roadmap lives in issue #6, and focused bug reports are welcome in GitHub Issues.

When reporting a photo-specific problem, please include the file format, camera model, expected behavior, and whether the file contains an embedded JPEG preview. Do not upload private photographs unless you are comfortable making them public.

For the ones who care how the shot was made—not just how it looks.