Howtos for The Rest of Us

← Backup

Backup

AI models are large - often 2–15 GB each - and re-downloading them is slow. A solid backup plan saves hours.

What to Back Up

• Model files (.safetensors, .bin, .pt) - irreplaceable or time-consuming to re-download
• Config files - workflow settings, custom node configs, environment variables
• Outputs - generated images, prompts, results (if you want to keep them)

What Not to Back Up

• Virtual environments (venv/) - recreate with pip install -r requirements.txt
• Git repos - re-clone instead
• Cache folders - HuggingFace cache, pip cache, etc.

Storage Tips

Keep models on a separate drive:

D:\AI_VAULT\models\
├── llm\              ← 10–50 GB each
├── diffusion\        ← 2–7 GB each
└── embeddings\       ← smaller

Avoid syncing AI_VAULT with OneDrive, Dropbox, or Google Drive - they’ll choke on multi-GB model files and may corrupt them.

Backup Strategy

For models:

• Copy to an external drive as-is (they’re just files)
• Note down model URLs or HuggingFace IDs in a text file so you know what you had

For configs:

• Keep them in a Git repo (small text files)
• Document custom node URLs in a list

Quick backup script (PowerShell):

robocopy D:\AI_VAULT F:\backup\AI_VAULT /MIR /R:2 /W:5

/MIR mirrors the folder structure. /R:2 /W:5 retries twice with a 5-second wait.