Safetensors Model Inspector

Inspect .safetensors models from a desktop GUI and CLI.

What It Does

• Detects architecture families and variants (Flux, SDXL/SD3, Wan, Hunyuan, Qwen, HiDream, LTX, Z-Image, Chroma, and more)
• Detects adapter type (LoRA, LyCORIS, LoHa, LoKr, DoRA, GLoRA)
• Extracts training metadata when present (steps, epochs, images, resolution, software, and related fields)
• Supports file or folder workflows (including recursive folder scanning)
• Supports .modelinfo key dumps for debugging and sharing

Repository Layout

• gui.py: GUI only
• inspect_model.py: model parsing, detection logic, data extraction, CLI
• requirements.txt: dependencies
• venv_create.bat: virtual environment bootstrap helper
• venv_activate.bat: activate helper

Setup

1. Create the virtual environment:

venv_create.bat

2. Activate:

venv_activate.bat

3. Run GUI:

py gui.py

4. Run CLI help:

py inspect_model.py --help

CLI Usage

Inspect one or more files

py inspect_model.py path\to\model1.safetensors path\to\model2.safetensors

Inspect folders

py inspect_model.py path\to\folder
py inspect_model.py path\to\folder --recursive

JSON output

py inspect_model.py path\to\folder --recursive --json

Write .modelinfo files

py inspect_model.py path\to\folder --recursive --write-modelinfo

Dump key/debug report text to console

py inspect_model.py path\to\folder --recursive --dump-keys

Optional alias fallback (filename tokens)

py inspect_model.py path\to\folder --recursive --allow-filename-alias-detection

GUI Walkthrough

Top Area (Input + Controls)

• Drag and drop files or folders into the drop zone
• Use Browse... or Browse Folder...
• Analyze processes queued inputs
• Settings controls visibility and behavior
• Minimize / Restore collapses or expands the top area for more workspace

Tab: Simple Cards

• Lightweight model cards
• Supports card selection, multi-select, and context menu actions

Tab: Detailed Cards

• Full card details with configured metadata visibility
• Supports card selection, multi-select, and context menu actions

• Supports specific LoRA formats like LoHa, LoKr, GLoRa
• Some fail sometimes (lycoris)

Tab: Data

• Sortable/resizable table
• Multi-select cells and copy via Ctrl+C
• Right-click actions (View Raw, Copy Selected Entries)
• Column visibility can be configured in settings

Tab: Raw

• Per-model raw .modelinfo text view
• View Raw context action jumps here for the selected model
• Ctrl+C copies the selected text, or the full raw content when no selection exists

Notes

• Folder drag/drop and folder browse both support recursive discovery of .safetensors.
• Filtering in the UI affects visibility and copy behavior (hidden rows are excluded from table copy).
• .modelinfo output is generated by shared backend logic in inspect_model.py.
• Filename alias detection is opt-in in Settings and can map filename tokens to fallback labels.
• Pony7 is treated as distinct from PDXL. The alias tokens pony7, ponyv7, and pony v7 map to Pony7.

Settings (Current)

General

• Filename Alias Detection: optional filename-token fallback for special labels
• Auto-minimize top section on Analyze
• Auto-analyze when files are added
• File add behavior:
  • Replace current input list
  • Append to current input list
• Default tab: Simple Cards, Detailed Cards, Data, or Raw

Visibility Groups

• Simple Cards: choose which data fields are shown
• Detailed Cards: choose which data fields are shown
• Data Columns: choose visible columns in the Data tab