- Only the first image of a batch was uploaded; the rest were silently
dropped. Every image is now uploaded (single images keep the plain
"<prefix>.<ext>" name, batches are numbered _001, _002, ...).
- Folder names containing an apostrophe broke the Drive search query.
- The placeholder json/credentials.json and token.json shipped with the
node produced confusing JSON errors; they're now detected and the node
says exactly which file to replace.
- Credentials and the Drive client are reused between runs, and folder IDs
are cached per path, so repeat saves skip the folder lookups (previously
every save re-read the token, rebuilt the client and searched each folder).
- Images under 5 MB upload in a single request instead of a resumable
upload session.
- Works with GPU tensors (.cpu() before .numpy()).
- Removed unused imports and the WEB_DIRECTORY that pointed at a
nonexistent folder; version 1.2.0 in pyproject.toml and __init__.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Lm4YDXhwj49hAA1RKS58Q
Claude