File Formats & Locations

Supported file formats, project structure, and where Recaster stores data on your system.

Supported Input Formats

Recaster accepts the following file formats as input for face swapping, enhancement, and upscaling operations.

Image Formats

Format	Extension	Notes
JPEG	`.jpg, .jpeg`	Most common. Lossy compression.
PNG	`.png`	Lossless. Recommended for source faces.
BMP	`.bmp`	Uncompressed bitmap. Large file sizes.
TIFF	`.tiff, .tif`	Lossless. Common in professional workflows.

Video Formats

Format	Extension	Notes
MP4	`.mp4`	Recommended. H.264/H.265 codec support.
AVI	`.avi`	Legacy format. Larger file sizes.
MOV	`.mov`	Apple QuickTime format. Full support.
MKV	`.mkv`	Matroska container. Multiple audio tracks supported.
WebM	`.webm`	Web-optimized format. VP8/VP9 codec.

Output Formats

Recaster produces the following output formats:

Images: .jpg and .png (configurable in settings)
Videos: .mp4 (H.264 codec with original audio preserved)

Audio Preservation

When processing videos, Recaster automatically detects and preserves the original audio track. The audio is extracted, the video frames are processed, and the audio is muxed back into the final output using ffmpeg.

Project Directory Structure

Each Recaster project is stored as a directory containing the following subdirectories. This structure mirrors the standard DeepFaceLab workspace layout for compatibility.

project-name/

src/— Source face images

dst/— Destination (target) face images

model/— Trained model files

merged/— Merged output frames

data_src.mp4— Source video file (if applicable)

data_dst.mp4— Destination video file (if applicable)

result.mp4— Final output video

DeepFaceLab Compatibility

Recaster projects use the same directory layout as DeepFaceLab workspaces. You can open existing DFL workspace folders directly in Recaster, and Recaster projects can be used with DFL command-line tools.

Settings & Data Locations

Recaster stores settings, license keys, training history, SSH keys, and cached models in a platform-specific application data directory.

macOS Paths

Settings Directory~/Library/Application Support/Recaster/

Settings File~/Library/Application Support/Recaster/settings.json

License Key~/Library/Application Support/Recaster/license.key

Training History~/Library/Application Support/Recaster/training_history.json

SSH Keys~/Library/Application Support/Recaster/ssh/

Model Cache~/Library/Application Support/Recaster/models/

Safety Audit Log~/Library/Application Support/Recaster/safety_audit.jsonl

Consent Audit Log~/Library/Application Support/Recaster/consent_audit.jsonl

Legacy Path Migration

Older versions of Recaster stored data in ~/.dfl-desktop/. This path is automatically migrated to the new location on first launch. No manual action is required.

Model Cache

AI models (face swap, enhancement, upscaling, classification) are downloaded from Hugging Face on first use and cached in the models/ subdirectory of your settings folder. Individual model files are typically 50-70 MB each.

To free disk space, you can safely delete the entire models directory. Models will be re-downloaded automatically the next time they are needed.

Settings File Structure

The settings.json file stores application preferences, consent state, and optional authentication tokens. Key fields include:

hf_token — Hugging Face authentication token for downloading gated models (e.g. SwinIR)
bipa_consent — BIPA consent state with timestamp and version
content_safety — Content safety configuration (admin-only)
gpu_compatibility_warning_dismissed — RTX 5000 warning suppression flag

Resetting Settings

If Recaster is not starting properly, deleting or renaming the settings.json file will reset all preferences to defaults. You will need to re-accept the BIPA consent dialog and reactivate your license key after a reset.

Was this page helpful?