FaceFusion 3.3.2 detailed settings Settings (intermediate level)

2025-08-14

FaceFusion has evolved significantly with the 3.3.2 update.
It is now more configurable than in previous versions,
FaceFusion 3.3.2 is now more configurable than in previous versions and can generate videos that meet your most sophisticated needs.

Basically, you can generate high quality face composite videos with the default settings.

This section describes detailed settings for those who wish to generate more advanced videos.
Please note that some settings may require more resources and significantly increase processing time,
This is important to keep in mind.

PROCESSORS section

  • face_swapper
    Enabled by default to allow face swapping.
    The core face change functionality provides for face replacement based on models such as inswapper_128.
  • deep_swapper (deep swapper)
    Alternative deep-fake model (not enabled by default)
    Default: iperov/elon_musk_224
    Selection: druuzil, edel, iperov, jen, mats, rumateus
    Example: --deep-swapper-model iperov/jackie_chan_224
  • face_debugger (face debugger)
    Useful for debugging during mask applications. Not required for normal operation. Displays the results of face keypoint detection for debugging purposes.
  • face_enhancer
    Used to enhance the clarity of a swiped face.
    Enhanced details of facial parts (eyes, skin, etc.).
    Model: gfpgan_1.4 Use the GFPGAN v1.4 model to enhance facial details (eyes, skin texture, etc.).
    Blend: 80 Degree of blending with the original image (0-100). The higher the value, the more pronounced the enhancement.
  • frame_enhancer
    Enhances the image quality. Shadow to performance
    Improves the overall clarity of video frames (super resolution).
    Model: span_kendata_x4 Super-resolution models (such as kendata and Real-ESRGAN) improve image clarity.
    Blend: 80 Also improves blending intensity.
  • age_modifier (age correction)
    Modifies the age of the character
  • expression_restorer (expression correction)
    The Expression Restorer module fixes problems with dull expressions after a face change.
    expression_restorer Intensity coefficient (0-100). Higher values are more expressive, but more prone to distortion
  • face_editor (face editor)
    Edits facial features such as eye size, mouth shape, etc. (if supported).
  • frame_colorizer (frame colorizer)
    Colorizes black and white videos and images.
  • lip_syncer (Lip Sync)
    Lip-sync, face swapping for audio-driven video (e.g. talking heads).

FACE SWAPPER MODEL section

  • FACE SWAPPER MODEL
    MODEL: New swapper HyperSwap 256 model, with twice the resolution and supposedly comparable to the Inswapper, has been released.
  • FACE SWAPPER PIXEL BOOST
    Pixel Boost: 256x256
    Enlarges the input image to this size for processing. Larger values produce sharper images, but also consume more video memory.

EXECUTION section

  • EXECUTION PROVIDERS
    Execution Providers: cuda,, sensorrtcpu
    The preferred compute backend to use is specified in the following order: sensorrt > cuda > cpu (if supported).
  • EXECUTION THREAD COUNT
    Number of threads:16 The number of multithreaded parallel operations is typically set according to the number of CPU cores.
  • EXECUTION QUEUE COUNT
    Number of queues:1 The number of task queues is usually left at 1.
  • DOWNLOAD PROVIDERS
    Download providers: github, huggingface
    Default source address used when downloading models.
  • video memory strategy
    Balance between fast frame processing and low VRAM usage.
    Video memory strategy: strict, default
    The fault tolerance policy tolerant means that graphics memory is automatically downgraded when graphics memory is full.
  • SYSTEM MEMORY LIMIT
    System memory limit: 0
    Upper limit of memory usage (MB). 0 means no limit

OUTPUT AUDIO ENCORDER section

Specifies the audio format of the generated video. Default is “flac”. (This section is displayed after the TARGET video is uploaded.)
(This section is displayed after TARGET video is uploaded.) Most media players use “aac”, so it is recommended to switch.

FACE SELECTOR MODE section

Set according to the number of faces in the target video.

  • REFERENCE
    If there are multiple faces in the video, select the face to be replaced from the faces in the reference image immediately below (REFERENCE FACE) and execute.
  • many
    If there are multiple faces in the video, replace all faces with one face (SOURCE).
  • one
    If there is only one face in the video.

FACE MASK TYPES section

  • FACE MASK TYPES
    Mixes and matches different face mask types.
    Default: box
  • FACE MASK BLUR
    Specifies the degree of blur applied to the box mask. Default: 0.3

FACE SELECTOR Section

  • FACE SELECTOR ORDER
    Specifies the order of the detected faces.
  • FACE SELECTOR GENDER
    Filters detected faces based on gender.
  • FACE SELECTOR RACE
    Filters detected faces based on race.

FACE MASKER section

  • FACE OCCLUDER MODEL
  • FACE PARSER MODEL
    Selects the model responsible for region masking.
  • FACE MASK PADDING TOP
    Applies top, right, bottom, and left padding to the box mask.

FACE DETECTOR ANGLES section

  • FACE DETECTOR ANGLES
    Specifies if the face angle in the TARGET video is 90, 180, or 270 degrees.
  • FACE DETECTOR SCORE
    ANGLESFilters the base of detected faces into a confidence score.