Research map / Image generation / Editing and style transfer
Face Adapter for Pre-Trained Diffusion Models with Fine-Grained ID and Attribute Control
Yue Han, Junwei Zhu, Keke He, Xu Chen, Yanhao Ge, Wei Li, Xiangtai Li, Jiangning Zhang, Chengjie Wang, Yong Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Adapting Area Predictor (background in spatial condition) (ablation) | Face-Adapter reenactment · VoxCeleb1 cross-identity | FID: 33.61 -> 31.18 | Improved |
| add Adapting Area Predictor (ablation) | Face-Adapter face swapping · VoxCeleb1 | FID: 33.97 -> 30.78 | Improved |
| add Adapting Area Predictor (ablation) | Face-Adapter face swapping · VoxCeleb1 | ID: 0.4530 -> 0.4688 | Improved |
| fine-tune CLIP for attribute features instead of freezing (ablation) | Face-Adapter reenactment · VoxCeleb1 cross-identity | FID: 33.09 -> 31.18 | Improved |
| fine-tune CLIP for attribute features instead of freezing (ablation) | Face-Adapter reenactment · VoxCeleb1 cross-identity | ID (CSIM): 0.6474 -> 0.6616 | Improved |
| fine-tune CLIP for attribute features instead of freezing (ablation) | Face-Adapter face swapping · VoxCeleb1 | ID: 0.4696 -> 0.4688 | No clear effect |
| add Adapting Area Predictor (ablation) | Face-Adapter swapping · VoxCeleb1 | Pose error: 0.0395 -> 0.0406 | No clear effect |
| use Face-Adapter instead of FADM on cross-identity reenactment | reenactment · VoxCeleb1 cross-identity | CSIM: 0.6152 -> 0.6429 | Improved |
| use landmark-based spatial control from D3DFR | Face-Adapter reenactment · VoxCeleb1 same-identity | Exp error: 2.70 -> 3.24 | Hurt |
| use Face-Adapter instead of diffusion DiffSwap for face swapping | face swapping · FaceForensics++ | ID retrieval: 19.16 -> 96.47 | Improved |
| use target identity tokens as negative prompt in face swapping inference | Face-Adapter face swapping · qualitative | identity similarity | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 3 on most shared results.
- HyperReenact — HyperReenact: One-Shot Reenactment via Jointly Learning to Refine and Retarget Faces (ICCV 2023) · better on 1 of 2 shared results
- PIRenderer — PIRenderer: Controllable Portrait Image Generation via Semantic Neural Rendering (ICCV 2021) · better on 1 of 2 shared results
- DiffSwap — DiffSwap: High-Fidelity and Controllable Face Swapping via 3D-Aware Masked Diffusion (CVPR 2023) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- High-Fidelity Diffusion Face Swapping with ID-Constrained Facial Conditioning (CVPR 2026 · similarity 0.92)
- DiffSwap: High-Fidelity and Controllable Face Swapping via 3D-Aware Masked Diffusion (CVPR 2023 · similarity 0.90)
- DynamicFace: High-Quality and Consistent Face Swapping for Image and Video using Composable 3D Facial Priors (ICCV 2025 · similarity 0.90)
- Smooth-Swap: A Simple Enhancement for Face-Swapping With Smoothness (CVPR 2022 · similarity 0.89)
- Learning Disentangled Representation for One-Shot Progressive Face Swapping (TPAMI 2024 · similarity 0.88)
- OSDFace: One-Step Diffusion Model for Face Restoration (CVPR 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
