FaceOff-FaceSwapper / README.md
Jds20001's picture
Fix ZeroGPU lifecycle + V2V guide-video conditioning (README.md)
253ec70 verified
|
Raw
History Blame Contribute Delete
2.53 kB

A newer version of the Gradio SDK is available: 6.20.0

Upgrade
metadata
title: FaceOff-FaceSwapper
emoji: 🎭
colorFrom: purple
colorTo: blue
sdk: gradio
sdk_version: 5.50.0
app_file: app.py
pinned: true
license: other
suggested_hardware: zero-a10g

FaceOff-FaceSwapper

Hugging Face Space for the V3 persistent-template head-swap workflow built on LTX-Video.

How it works

  1. Upload a guide video (the body/performance you want to keep)
  2. Upload a reference face image (the identity to transfer)
  3. Write or paste a text prompt in the structured format below
  4. Click Generate

The reference face is composited into a green chroma side-strip on every frame of the guide video, giving the model continuous identity conditioning. The composed video is then re-rendered (video-to-video) with the swapped head. The strip is cropped from the final output automatically.

Prompt format

head_swap:
FACE: <detailed identity description from the reference image>
ACTION: <performance description from the guide video — body, clothing, movement only>

Example:

head_swap:
FACE: Male, fair skin, approximately 25-30 years old, short light brown hair,
blue eyes, clean-shaven, athletic build, wearing a navy blue t-shirt.

ACTION: A person walks confidently toward the camera in an outdoor plaza,
arms crossed, smiling.

Models used

Component Source
Full pipeline (transformer, VAE, text encoder, scheduler) Lightricks/LTX-Video-0.9.8-13B-distilled
BFS head-swap LoRA Alissonerdx/BFS-Best-Face-Swap-Video (ltx-2/head_swap_v2_multimodes.safetensors)
Auto-describe Salesforce/blip2-opt-2.7b
Face alignment InsightFace buffalo_l (CPU)
Optional enhancement GFPGAN v1.4

Logs

Container logs (runtime output):

curl -N \
     -H "Authorization: Bearer $HF_TOKEN" \
     "https://api.hf.space/v1/Jds20001/FaceOff-FaceSwapper/logs/run"

Build logs (Docker build output):

curl -N \
     -H "Authorization: Bearer $HF_TOKEN" \
     "https://api.hf.space/v1/Jds20001/FaceOff-FaceSwapper/logs/build"

Ethical use

This tool is intended for filmmakers, VFX artists, and researchers. You must have explicit consent from any person whose likeness you process. See the full model card for the complete terms.