Research map / Image generation / Editing and style transfer
Arc2Face: A Foundation Model for ID-Consistent Human Faces
Foivos Paraperas Papantoniou, Alexandros Lattas, Stylianos Moschoglou, Jiankang Deng, Bernhard Kainz, Stefanos Zafeiriou
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 |
|---|---|---|---|
| train ID-to-face model on WebFace42M instead of FFHQ | ID-conditioned latent diffusion trained from scratch · Synth-500 and AgeDB | ID similarity | Improved |
| train on 10% of WebFace42M (about 4M images) instead of full | ID-conditioned latent diffusion trained from scratch · Synth-500 and AgeDB | ID similarity | No clear effect |
| project ArcFace through the fine-tuned CLIP text encoder instead of a 4-layer MLP | Arc2Face · Synth-500 and AgeDB | ID similarity | Improved |
| condition only on ID embeddings instead of ID plus text or CLIP image features | Arc2Face · Synth-500 (prompt 'photo of a person') | LPIPS diversity: 0.410 (Photomaker) -> 0.506 | Improved |
| use ID embeddings without landmark conditioning instead of InstantID's landmarks | Arc2Face · Synth-500 | pose l2 diversity: 0.059 -> 0.317 | Improved |
| train FR model on Arc2Face synthetic data instead of DCFace data (0.5M) | IR-SE-50 + AdaFace · 5-dataset average | verification accuracy: 89.56 -> 91.73 | Improved |
| scale synthetic FR training data from 0.5M to 1.2M images | Arc2Face · 5-dataset average | verification accuracy: 91.73 -> 93.14 | Improved |
| compare synthetic Arc2Face data with real CASIA-WebFace for FR training | IR-SE-50 + AdaFace · 5-dataset average | verification accuracy: 94.62 (real) vs 91.73 | Hurt |
| reduce ArcFace embeddings to fewer PCA components | Arc2Face · WebFace42M ID embeddings | facial fidelity | Hurt |
| average ArcFace embeddings of multiple images of same person | Arc2Face · 400 AgeDB IDs | ID similarity | Improved |
| check memorization via closest training image | Arc2Face · Synth-500 unseen IDs | ArcFace similarity: 0.37 to closest train image vs 0.74 to input | No clear effect |
Later papers that compared with this method
8 later accepted papers report Arc2Face (10K) as a baseline in their results tables; 7 of them report a better number on most of the results they share with it.
- Non-Colliding Biometric Identities for Digital Entities: Geometry, Capacity, and Million-Scale Virtual Identity Provisioning (NeurIPS 2026 · better on 4 of 4 shared results)
- IDperturb: Enhancing Variation in Synthetic Face Generation via Angular Perturbations (CVPR 2026 · better on 1 of 1 shared result)
- Data Synthesis with Diverse Styles for Face Recognition via 3DMM-Guided Diffusion (CVPR 2025 · better on 6 of 6 shared results)
- UIFace: Unleashing Inherent Model Capabilities to Enhance Intra-Class Diversity in Synthetic Face Recognition (ICLR 2025 · better on 4 of 4 shared results)
- ControlFace: Harnessing Facial Parametric Control for Face Rigging (CVPR 2025 · better on 2 of 3 shared results)
- VIGFace: Virtual Identity Generation for Privacy-Free Face Recognition Dataset (ICCV 2025 · better on 2 of 2 shared results)
- HyperLoRA: Parameter-Efficient Adaptive Generation for Portrait Synthesis (CVPR 2025 · better on 1 of 1 shared result)
- Vec2Face: Scaling Face Dataset Generation with Loosely Constrained Vectors (ICLR 2025 · better on 0 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- DCFace — DCFace: Synthetic Face Generation With Dual Condition Diffusion Model (CVPR 2023) · better on 3 of 3 shared results
- Photomaker — PhotoMaker: Customizing Realistic Human Photos via Stacked ID Embedding (CVPR 2024) · better on 2 of 2 shared results
- FastComposer — FastComposer: Tuning-Free Multi-subject Image Generation with Localized Attention (IJCV 2024) · better on 2 of 2 shared results
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
- Vec2Face: Scaling Face Dataset Generation with Loosely Constrained Vectors (ICLR 2025 · similarity 0.90)
- $\text{ID}^3$: Identity-Preserving-yet-Diversified Diffusion Models for Synthetic Face Recognition (NeurIPS 2024 · similarity 0.88)
- HyperFace: Generating Synthetic Face Recognition Datasets by Exploring Face Embedding Hypersphere (ICLR 2025 · similarity 0.88)
- UIFace: Unleashing Inherent Model Capabilities to Enhance Intra-Class Diversity in Synthetic Face Recognition (ICLR 2025 · similarity 0.88)
- FaceA-Net: Facial Attribute-Driven ID Preserving Image Generation Network (AAAI 2025 · similarity 0.86)
- FaithfulFaces: Pose-Faithful Facial Identity Preservation for Text-to-Video Generation (NeurIPS 2026 · similarity 0.86)
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.
