BunnyCard

BunnyCard is a local media ingest tool for camera cards and delivery drives.

It copies videos and photos into separate libraries, renames files with a stable device_date_profile_tag_material pattern, verifies every copied file with SHA-256, and writes audit logs for each run.

Run the GUI

cd /Users/davinci/Documents/GitHub/BunnyCard
swift run BunnyCard

Use 生成预览 to build a no-write table showing each source file, category, target folder, new filename, size, and status. Select a row to inspect the full source and target paths plus a local thumbnail preview. Use 开始复制 only after the preview looks right.

The GUI shows mounted volume cards first. It filters out the system disk and highlights volumes with camera-card structures such as DCIM, PRIVATE, and M4ROOT.

Default Paths

• Source card: /Volumes/TF256GB_San
• Video library: /Volumes/PS2000/MediaLibrary
• Photo library: /Volumes/StudioDisk/lightroom
• Audit logs: /Volumes/PS2000/_DuplicateAudit

CLI

Scan a card:

python3 src/bunnycard_core.py scan --source /Volumes/TF256GB_San

Preview an import:

python3 src/bunnycard_core.py plan-card \
  --source /Volumes/TF256GB_San \
  --video-library /Volumes/PS2000/MediaLibrary \
  --photo-library /Volumes/StudioDisk/lightroom \
  --audit-root /Volumes/PS2000/_DuplicateAudit \
  --language zh \
  --layout project_tag \
  --project RCJ2026 \
  --thumbnails \
  --thumbnail-limit 48

Run the import for real while keeping source files on the card:

python3 src/bunnycard_core.py import-card \
  --source /Volumes/TF256GB_San \
  --video-library /Volumes/PS2000/MediaLibrary \
  --photo-library /Volumes/StudioDisk/lightroom \
  --audit-root /Volumes/PS2000/_DuplicateAudit \
  --backup-library /Volumes/PS2000/CardBackups \
  --language zh \
  --layout project_tag \
  --project RCJ2026 \
  --delete-sidecars

Add --delete-source only when the card should be cleared after verified copy. When --backup-library is set, source deletion requires both the organized copy and the original-structure backup to verify.

Prepare an RCJ delivery drive:

python3 src/bunnycard_core.py prepare-rcj \
  --media-library /Volumes/PS2000/MediaLibrary \
  --target /Volumes/Netac_256GB/RCJ_Delivery_20260628 \
  --audit-root /Volumes/PS2000/_DuplicateAudit

Classification

The built-in rule classifier handles common paths and names:

• DJI is treated as a brand, not as automatic 航拍.
• DJI Osmo/Pocket/Action media goes to 手持素材 unless visual/path evidence says otherwise.
• DJI drone model/path evidence goes to 航拍; D-Log/D-Log M files include dlogm.
• RCJ/RoboCup material goes to RoboCup.
• Robot setup material goes to 机器人调试.
• Graduation and sports field names get their own categories.

Use --language en or the GUI language segmented control for English folder and filename tags.

For local visual classification, start Ollama and enable Use local Ollama in the GUI, or pass --use-ollama --ollama-model llava in the CLI. The engine samples one frame with ffmpeg; if Ollama fails, it falls back to rules. HTTP Error 404 from Ollama means the selected model is not installed. The GUI lists local models and warns when no vision-capable model is detected.

Audit Logs

Each run creates a folder like:

/Volumes/PS2000/_DuplicateAudit/bunnycard_YYYYMMDD_HHMMSS_CARDNAME

The useful files are:

• scan_manifest.csv
• copy_log.csv
• audio_sidecar_copy_log.csv
• backup_copy_log.csv
• source_delete_log.csv
• sidecar_delete_log.csv
• summary.json

TODO

• Proxy/transcode generation for editors.
• Face clustering / face-based grouping.
• System notification hooks after long imports.
• Export a compact human-readable PDF/HTML report.