Processing Clips (Wizard)
The wizard is the main way to process footage. It guides you through scanning, reviewing, and running inference interactively.
Starting the Wizard
corridorkey wizard /path/to/clips
If you omit the path, the wizard will prompt for it.
On Windows you can also drag a clips folder onto the CorridorKey - Drop Clips Here.bat shortcut on your Desktop.
Step 1: Organise (if needed)
Before showing the clip table, the wizard scans for unorganised content - loose video files or folders that contain raw footage without the expected Input/ subfolder. If it finds any, it shows a summary and asks whether to restructure:
2 loose video file(s) found:
- /path/to/clips/actor_jump.mp4
- /path/to/clips/product_spin.mp4
These will be restructured into clip folders with Input/ and empty AlphaHint/ subdirectories.
Organise now? [y/n] (n):
Choosing y moves each video into its own {stem}/Input.ext folder and creates empty AlphaHint/ and VideoMamaMaskHint/ subdirectories. The wizard then re-scans and shows the updated clip table.
If everything is already structured, this step is skipped silently.
See Preparing clips for details on what the organiser handles.
Step 2: The Clip State Table
The wizard shows a table of every clip it found and its current state:
Clips in /path/to/clips
Clip State Input Alpha Error
actor_jump EXTRACTING 0 0
my_shot READY 48 48
another_shot RAW 60 -
The state tells you what the pipeline will do with each clip:
| State | Meaning |
|---|---|
| READY | Has input frames and alpha hint. Will be processed when you choose [i]. |
| EXTRACTING | Source video detected. Frames will be extracted automatically before inference. |
| COMPLETE | Already processed. Will be skipped. |
| RAW | Has input frames but no alpha hint. Needs an alpha generator package. |
| MASKED | Has a VideoMaMa mask hint. Needs VideoMaMa to convert it. |
| ERROR | A previous step failed. See the Error column for details. |
READY and EXTRACTING clips are both processed when you choose [i]. RAW and MASKED clips are skipped with a note.
Step 3: The Action Menu
[i] Process 1 READY, 1 to extract clip(s)
[r] Re-scan directory
[q] Quit
Choose i to run inference (extracting any EXTRACTING clips first), r to re-scan after adding or changing clips, or q to quit.
Step 4: Inference Settings
Before processing starts, the wizard shows the current settings and asks whether to use them:
Setting Value
input_is_linear False
despill_strength 1.0
auto_despeckle True
despeckle_size 400
refiner_scale 1.0
Use these settings? [y/n] (y):
Press Enter to accept the defaults. Enter n to adjust each setting interactively.
What each setting does
input_is_linear - Set to True if your frames are linear light (e.g. EXR from a camera or renderer). Leave False for standard sRGB footage (PNG, JPEG from a camera).
despill_strength - Controls how aggressively green colour contamination is removed from the subject. 1.0 is maximum. Reduce if skin tones look too magenta.
auto_despeckle - Removes small isolated holes and islands in the matte automatically. Leave enabled for most footage.
despeckle_size - The maximum size (in pixels) of a matte artefact to remove. Increase if small holes remain; decrease if fine detail like hair is being removed.
refiner_scale - Controls the edge refinement pass. 1.0 is the default. Reduce toward 0.0 to speed up processing at the cost of edge quality.
Model Loading
The first time you run inference in a session, the wizard loads the model and compiles GPU kernels. This takes about one minute and shows a spinner:
Loading model (first run compiles kernels, ~1 min)...
Subsequent clips in the same session process immediately because the model stays loaded.
Extraction
EXTRACTING clips are unpacked before inference runs. The wizard shows progress for each:
Extracting actor_jump (0 frames)
After extraction the clip advances to READY (if an alpha video was also extracted) or RAW (if no alpha was found). RAW clips after extraction are skipped unless an alpha generator is installed.
Progress
Each clip shows a progress bar while processing:
Processing my_shot (48 frames)
Done: 48/48 frames 18.3s (2.62 fps)
After all clips finish, the wizard returns to the main loop so you can re-scan or process more clips.
Batch Processing (Non-Interactive)
For scripted or automated use, corridorkey process runs without any prompts:
corridorkey process /path/to/clips --despill 0.9 --fg-format exr
See Processing commands for all available flags.