FaceFusion 3.3.2

FACE ENHANCER MODEL

Choose an AI model to enhance face quality and details. Recommended: gfpgan_1.4 or restoreformer_plus_plus for best results.

FACE ENHANCER BLEND

Control the blend between enhanced and original face. 100 = full enhancement, 0 = original.

0 100

FACE SWAPPER MODEL

Choose the AI model for face swapping. inswapper_128 and blendswap_256 offer the best quality.

FACE SWAPPER PIXEL BOOST

Higher resolution produces better quality but requires more processing time and memory. Use 1024x1024 for best results.

EXECUTION PROVIDERS

Select hardware acceleration. CUDAExecutionProvider for NVIDIA GPUs, CoreMLExecutionProvider for Apple Silicon, CPUExecutionProvider as fallback.

cpu

EXECUTION THREAD COUNT

Number of parallel processing threads. Higher = faster but uses more CPU/GPU. Set to CPU core count for optimal performance.

1 32

EXECUTION QUEUE COUNT

Frames each thread processes before returning. Higher = better GPU utilization but more VRAM needed. 1-2 recommended.

1 4

DOWNLOAD PROVIDERS

github huggingface

VIDEO MEMORY STRATEGY

Balance between speed and VRAM usage. Strict = low memory, Moderate = balanced, Tolerant = faster but uses more VRAM.

SYSTEM MEMORY LIMIT

0 128

SOURCE

TARGET

OUTPUT PATH

Specify the output directory or file path where the processed result will be saved.

OUTPUT

LOG LEVEL

TERMINAL

UI WORKFLOW

PREVIEW

FACE SELECTOR MODE

Choose how to select faces: Reference (track specific face), One (first detected), Many (all faces), or filters by age/gender/race.

REFERENCE FACE

FACE SELECTOR ORDER

FACE SELECTOR GENDER

FACE SELECTOR RACE

[

0,
100

]

REFERENCE FACE DISTANCE

0 1

FACE OCCLUDER MODEL

Model to detect occlusions (glasses, hands, hair covering face). Helps create better masks.

FACE PARSER MODEL

Model to segment face regions (eyes, nose, mouth, etc). Used for precise region masking.

FACE MASK TYPES

Masking methods: Box = simple rectangle, Occlusion = avoid covered areas, Region = specific face parts. Combine for best results.

box occlusion area region

FACE MASK BLUR

Blur amount on mask edges. Higher = smoother blend with background, more natural transitions. 0.3-0.5 recommended.

0 1

FACE MASK PADDING TOP

0 100

FACE MASK PADDING RIGHT

0 100

FACE MASK PADDING BOTTOM

0 100

FACE MASK PADDING LEFT

0 100

FACE DETECTOR MODEL

AI model for detecting faces in images/videos. yolo_face is recommended for best accuracy and speed.

FACE DETECTOR SIZE

Detection resolution - larger sizes detect faces better but slower. 640x640 balances speed and accuracy.

FACE DETECTOR ANGLES

Detect faces at different angles (rotation). Enable more angles to find tilted/rotated faces, but slower processing.

0 90 180 270

FACE DETECTOR SCORE

Confidence threshold for face detection (0-1). Higher values = stricter detection, fewer false positives. Default 0.5 works well.

0 1

FACE LANDMARKER MODEL

Model to detect facial landmarks (eye corners, nose tip, mouth points). Essential for accurate face alignment and swapping.

FACE LANDMARKER SCORE

Confidence threshold for landmark detection (0-1). Higher = more accurate landmarks required. 0.5 is generally good.

0 1

OPTIONS

keep-temp