facefusion-extras

Custom model mirror for FaceFusion Đợt 3–6.

This repo hosts ONNX weights that the upstream facefusion/facefusion-assets repo does not (yet) ship: SOTA 2024–2026 backends used by the new processors landing in subsequent PRs of ULTRA_ROADMAP.

Layout

facefusion-extras/
β”œβ”€β”€ README.md
β”œβ”€β”€ frame_interpolator/         # Đợt 1.A3 β€” RIFE 4.26 ONNX
β”œβ”€β”€ face_swapper_extras/        # Đợt 1.A1 β€” inswapper_512_live (if license-cleared)
β”œβ”€β”€ lip_syncer_extras/          # Đợt 3.A2 β€” LatentSync, MuseTalk v1.5
β”œβ”€β”€ image_restorer/             # Đợt 3.C β€” HYPIR / SUPIR-Face (opt-in research)
β”œβ”€β”€ portrait_animator/          # Đợt 4.D1 β€” LivePortrait v2 / AniPortrait
β”œβ”€β”€ temporal_stabilizer/        # Đợt 4.D2 β€” StableAnimator / FlowVid
β”œβ”€β”€ voice_cloner/               # Đợt 6.F β€” F5-TTS / OpenVoice v2
└── hashes/                     # SHA256 hash files for verification

Each .onnx (or .safetensors) weight has a sibling .hash file that FaceFusion uses to verify integrity at download time.

Usage

Set environment variable before running FaceFusion:

# Linux / macOS
export FACEFUSION_HF_NAMESPACE=ngoqquyen

# Windows cmd
set FACEFUSION_HF_NAMESPACE=ngoqquyen

# Windows PowerShell
$env:FACEFUSION_HF_NAMESPACE = "ngoqquyen"

When the new processors (frame_interpolator, portrait_animator, …) request weights with base_name = facefusion-extras, FaceFusion's download resolver substitutes the upstream facefusion/ namespace with ngoqquyen/, hitting this repo instead.

License

Each weight retains its original license. See LICENSE.<weight>.txt files in each subdirectory. By using this mirror you agree to comply with the upstream license of every weight you download.

This mirror is provided as-is, without warranty.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support