A Streamlit app to:

• Scan a dataset of PNG/JPG images (now with subfolder support)
• Extract prompts from PNG metadata (ComfyUI + common formats)
• Edit prompts in a table view
• Export/import as JSONL with portable relative paths
• Bulk actions (select/delete/export and add prefix/suffix)

• Recursive rescan: include subfolders under your dataset directory
• Portable JSONL:
  • Manifest header: {"__manifest__":{"base_dir":"./dataset"}}
  • Each entry rel_path is base_dir/dataset_filename (e.g. ./dataset/sub/img.png)
  • No absolute paths in JSONL
• Prompt extraction from:
  • PNG "parameters" text
  • ComfyUI "workflow" or "prompt" JSON (best-effort heuristics)
• Thumbnails and prompt extraction cached for speed
• Bulk actions:
  • Select on current page
  • Clear selection
  • Delete selected (session only)
  • Export selected as JSONL
  • Add prefix/suffix to selected prompts

python -m venv .venv
source .venv/bin/activate   # Windows: .venv\Scripts\activate
pip install -r requirements.txt

From project root:

python -m streamlit run app/app.py

Ensure app/init.py exists (can be empty).

• Set “Dataset Directory” in the sidebar (prefer a relative path like ./dataset)
• Click “Rescan Dataset Directory” (enable “Include subfolders” for recursive - scan)
• Edit prompts in the table
• Use bulk actions for selection, delete, export, prefix/suffix
• Save the entire session as JSONL via “Save Data”

First line is a manifest:

{"__manifest__":{"base_dir":"./dataset"}}

Each subsequent line is an item:

{
  "id": "...sha256...",
  "original_name": "image.png",
  "dataset_filename": "sub/image.png",
  "rel_path": "./dataset/sub/image.png",
  "prompt": "a sample prompt",
  "modified": false,
  "source": "rescanned_dataset"
}

To load on another machine:

• Place the dataset folder relative to the project root as in base_dir
• Load the JSONL (the app reconstructs absolute paths from rel_path

• Deletion from dataset directory is only performed for items added via upload (to avoid accidental removal of external files). Adjust in app/app.py if needed.
• Caching: thumbnails and prompt extraction are cached by Streamlit (st.cache_data).

• Graph-aware ComfyUI prompt extraction
• Sidecar save of edits alongside images
• CSV export
• Tests (pytest) and sample PNGs