FaceFusion AI Face Swap Guide: Settings, Masks, and Advanced Techniques

FaceFusion is a new generation AI face-swapping tool that runs with one click, requires no complicated installation, and supports occlusion removal, high-definition enhancement, and cartoon face replacement. It works across mainstream GPUs from Nvidia and AMD. The latest version adds new models and high-resolution algorithms, improving the quality of both image and video swaps, and introduces three mask modes that solve blending issues and flickering when faces are partially obstructed. This guide walks through the key parameters of the Gradio interface so you can achieve reliable face swaps.

Before starting, you need the FaceFusion open-source project, available on GitHub. For installation steps, refer to the previous tutorial in this series. This article focuses on how each parameter in the Gradio interface affects the final result, so that you can adjust settings with confidence.

Several settings matter most. FaceSwap must be enabled, otherwise no swap will occur. FaceDebug is only used for debugging masks and should be off for normal runs. Face Enhancer improves clarity and works well with the default GFPGAN 1.4 model. Frame Enhancer boosts overall image quality but significantly slows processing, so it is only recommended on RTX 4090-class GPUs. The default InSwapper_128 swap model works well in most scenarios. For Execution Provider, choose CPU for AMD or integrated graphics and CUDA for Nvidia cards. Thread count can be raised if you have 8GB or more of VRAM, and queue count is only needed for batch scripts. Under Options, enabling Skip-download is recommended once models are cached locally, while Keep-temp should stay off to save disk space and Skip-audio should stay off to preserve sound. To swap a single face, upload a clear front-facing photo, select the target image, click Start, and download the result. For sharper output, enable Face Enhancer and run the swap again.

For videos, select the source file and click the detected face to preview the result. The combination of Face Enhancer, the InSwapper_128 model, and GFPGAN 1.4 is a reliable default. Thread count can be increased based on available VRAM, and reducing it to around 12 can help if processing is slow. Enabling Keep original frame rate and Skip-download is recommended. Face selection defaults to detecting faces from left to right and replacing the first one; the Many option replaces all detected faces, while One replaces only the first. The Reference Face Distance parameter usually works at its default of 0.6. For videos with large head angles or side profiles, increasing it up to 1.5 allows more faces to be matched and swapped.

When faces are partly covered by objects, choosing the right mask mode is essential. Enable FaceDebug while adjusting masks so you can preview them. The default Box mode draws a green rectangle around the face. For makeup or partial obstructions, the Occlusion mode follows the actual contours of the obstruction. Face Mask Blur, defaulting to 0.3, controls transparency of the mask; opaque objects can use values above 0.5, while transparent items like glasses often look more natural with lower values. The Region mode lets you choose specific facial parts such as eyebrows, nose, eyes, or mouth to include or exclude from replacement, which works well for talking scenes. With Occlusion or Region modes, the four padding parameters (top, right, bottom, left) fine-tune the swapped area. In Box mode, these padding values control the size of the square mask. If subtitles are blurred by the mask, adjust the bottom padding until the red box no longer covers them. Always turn off FaceDebug before running the final swap.

For high-resolution footage, especially close-up shots, the Face Detector Size setting can be raised up to 1024 pixels for more accurate detection. When Face Enhancer is enabled, the CodeFormer algorithm can replace GFPGAN 1.4 in close-ups, producing more realistic skin texture.

Cartoon faces require a different setup. Choose the simswap_256 model instead of InSwapper_128 and enable Face Enhancer. In this configuration, the eyes, nose, and mouth are replaced while the rest of the image remains unchanged, producing a convincing stylized result.

This guide has walked through the full range of FaceFusion features, from installation and single-image swaps to video processing, mask configuration, high-definition handling, and cartoon face replacement. Mastering these parameters allows you to produce consistent and high-quality AI face swaps across a wide variety of content.