ImageArc

Open-source, cross-platform (Linux/Windows/macOS) AI photo captioning and IPTC/XMP metadata tool for photographers. Pure Go, no CGO, single static binary.

ImageArc never rewrites your image pixels. It calls out to ExifTool to write standard IPTC/XMP metadata: in-file for JPEG/TIFF, and Lightroom-style .xmp sidecars for RAW (CR2/CR3/NEF/ARW/DNG/...).

Features

• AI captioning via local Ollama (llava, qwen2.5vl, qwen3-vl) — fully implemented, works offline.
• Provider interface ready for Anthropic, OpenAI, Gemini, and any OpenAI-compatible endpoint (stubs included, easy to fill in).
• Non-destructive metadata writes through ExifTool: caption, keywords, byline, location, headline, date — mapped to both IPTC and XMP fields.
• Recursive folder pipeline with a configurable worker pool and live progress.
• CSV catalog builder, with or without AI (--no-ai just reads existing IPTC).
• Per-client JSON templates (byline, location, keyword prefixes, caption prefix).
• Embedded dark-theme web UI: batch runner with a live activity log (SSE), metadata editor, and catalog builder — no external CDNs, one binary.
• No lock-in: everything lands in plain CSV, JSON, and XMP/IPTC.

Install

Requirements:

• Go 1.22+ (to build from source)
• ExifTool on your PATH
  • Linux: sudo apt install libimage-exiftool-perl (or your distro's package)
  • macOS: brew install exiftool
  • Windows: download from exiftool.org and add to PATH
• Ollama for local AI captioning, plus a vision model:

  ollama pull llava
  # or: ollama pull qwen2.5vl / ollama pull qwen3-vl
  

Build

go build -o imagearc ./cmd/imagearc

Cross-compile

GOOS=linux   GOARCH=amd64 go build -o dist/imagearc-linux-amd64     ./cmd/imagearc
GOOS=windows GOARCH=amd64 go build -o dist/imagearc-windows-amd64.exe ./cmd/imagearc
GOOS=darwin  GOARCH=arm64 go build -o dist/imagearc-macos-arm64     ./cmd/imagearc

Usage

Caption a folder

imagearc caption ./photos --provider ollama --model llava --recurse

Dry run (caption only, no metadata written):

imagearc caption ./photos --dry-run

With a client template and a custom worker count:

imagearc caption ./photos --recurse --template client.json --workers 4

client.json:

{
  "name": "Acme Wedding Co",
  "byline": "Jane Doe Photography",
  "location": "Austin, TX",
  "keywordsPrefix": ["wedding", "acme-co"],
  "captionPrefix": "Acme Wedding Co:"
}

Build a catalog

imagearc catalog ./photos --recurse -o catalog.csv

--no-ai is the default and only mode today: catalog reads existing IPTC/XMP via ExifTool and writes CSV — no network, no AI calls.

Serve the web UI

imagearc serve --addr localhost:8733

Opens your browser to a dark-themed UI for batch captioning (with a live activity log), a metadata editor, and catalog building.

Version

imagearc version

How metadata is written

ImageArc maps fields to both IPTC and XMP so readers using either standard pick them up:

Field	IPTC	XMP
Caption	Caption-Abstract	dc:Description
Keywords	Keywords	dc:Subject
Byline	By-line	dc:Creator
Location	Sub-location	iptcCore:Location
Headline	Headline	—
Date	DateCreated	—

For RAW files, a <file>.xmp sidecar is created (if missing) and then updated via ExifTool — the original RAW file is never touched. For JPEG/TIFF, ExifTool writes in place with -overwrite_original (a backup is skipped since nothing but metadata changes, and the pixel data is untouched by ExifTool tag writes).

Use with Shotwell / other photo managers

ImageArc writes standard IPTC/XMP, so photo managers that read embedded metadata (Shotwell, digiKam, Lightroom, darktable) pick up the AI captions and keywords as tags — no plugin needed. A right-click "Caption with ImageArc" action for GNOME Files/Nemo/Caja and the Shotwell workflow are in integrations/.

No lock-in

• Catalogs are plain CSV.
• Templates are plain JSON.
• Metadata lives in your files/sidecars as standard IPTC/XMP, readable by Lightroom, digiKam, darktable, or any other tool.

Repositories

Mirrored on both forges — clone from whichever you prefer:

• Codeberg: https://codeberg.org/YR-Design/imagearc
• GitHub: https://github.com/robouden/imagearc

Origin

The concept behind ImageArc lived in my mind for years, discussed at length with fellow photographers like Paul Disario and Michael Goldberg. It was finally woken up and set in motion by Peter Blakely's Mac-only FotoArch/MetaEdit — ImageArc grew out of the wish for a free, open-source, cross-platform take on that idea.

Credits

Offline reverse geocoding (GPS → city/country) uses the GeoNames cities15000 dataset, licensed CC BY 4.0.

License

CC0 1.0 Universal — public domain dedication. No rights reserved.