STATE-AWARE · TRACEABLE · AUTHORIZED
DeepFake Redteam AttackerDual-Branch Generation Pipeline · v0.2
A research-oriented framework integrating face swapping, source-aware facial editing, modular image refinement, and detector-based evaluation.
Precomputed Demo：这里展示已保存的历史样本，选择与按钮仅切换静态结果。Detector prediction is not equivalent to attack success.
01 / INTERACTIVE ARCHITECTURE
Dual-Branch Pipeline Architecture
Original final architecture · Click or Tab / Enter to explore modules. This diagram documents the research system; the public website serves static assets only.
Focus a module to see its role.
02 / PRECOMPUTED DEMO
Face Swapping Demo
Source Identity + Target → one of three stored swapping results.
Only exact recorded combinations are available. Changing the Generator / Action selects its corresponding preset; manually choosing a different input combination may have no stored sample.
Choose a preset, then View Result
Source, Generator Output and Final Candidate will be shown for the exact stored selection.
03 / PRECOMPUTED DEMO
Face Editing Demo
Preset Source + recorded action → stored LivePortrait result.
Only exact recorded combinations are available. Changing the Generator / Action selects its corresponding preset; manually choosing a different input combination may have no stored sample.
Choose a preset, then View Result
Source, Generator Output and Final Candidate will be shown for the exact stored selection.
04 / REFERENCES
Generation Methods & Official Sources
References do not grant redistribution rights to third-party code or weights. Neither is bundled here.
Responsible Use
For authorized research, robustness evaluation, and transparent demonstration. Do not impersonate people, mislead viewers, or use these samples without applicable consent. Public display permission does not create a blanket license to redistribute the images.