pith. machine review for the scientific record. sign in

arxiv: 2604.14541 · v1 · submitted 2026-04-16 · 💻 cs.CV

Recognition: unknown

Giving Faces Their Feelings Back: Explicit Emotion Control for Feedforward Single-Image 3D Head Avatars

Yicheng Gong , Jiawei Zhang , Liqiang Liu , Yanwen Wang , Lei Chu , Jiahao Li , Hao Pan , Hao Zhu
show 1 more author
Yan Lu
Authors on Pith no claims yet

Pith reviewed 2026-05-10 11:45 UTC · model grok-4.3

classification 💻 cs.CV
keywords 3D head avatarsemotion controlsingle-image reconstructionfeedforward networksdisentangled manipulationfacial animationemotion transfermodulation mechanism
0
0 comments X

The pith

Emotion can be treated as an independent control signal in single-image 3D head avatars.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper establishes a method to control emotions separately in 3D head avatars built from one photo, rather than letting them stay mixed with shape or appearance. It adds this control to existing feed-forward models through a dual-path modulation that adjusts geometry via emotion-conditioned normalization and appearance via identity-aware emotion cues. A supporting dataset is built by transferring aligned emotional dynamics across identities to keep timing and consistency intact. This keeps the original reconstruction quality while allowing users to transfer emotions, manipulate them apart from speech movements, and interpolate between states smoothly.

Core claim

We present a framework for explicit emotion control in feed-forward, single-image 3D head avatar reconstruction. Unlike existing pipelines where emotion is implicitly entangled with geometry or appearance, we treat emotion as a first-class control signal that can be manipulated independently and consistently across identities. Our method injects emotion into existing feed-forward architectures via a dual-path modulation mechanism without modifying their core design. Geometry modulation performs emotion-conditioned normalization in the original parametric space, disentangling emotional state from speech-driven articulation, while appearance modulation captures identity-aware, emotiondependent

What carries the argument

Dual-path modulation mechanism that separates geometry modulation (emotion-conditioned normalization in parametric space) from appearance modulation (identity-aware emotion-dependent cues).

If this is right

  • Existing feed-forward 3D head avatar architectures gain emotion control without changes to their core design.
  • Emotion transfer becomes controllable and consistent across different identities.
  • Emotional state can be disentangled from speech-driven articulation for separate manipulation.
  • Smooth interpolation between emotional states is supported while preserving reconstruction fidelity.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • This separation could support real-time emotion editing in virtual meetings or games using single photos.
  • The dataset construction technique might apply to other disentanglement tasks like age or lighting control.
  • Extending the modulation to video inputs could enable dynamic emotion sequences beyond static images.

Load-bearing premise

That emotional dynamics can be transferred and aligned across different identities to create a time-synchronized dataset without artifacts or identity leakage.

What would settle it

Apply the same emotion sequence to multiple identities in the dataset and check whether the outputs show consistent timing, no visible artifacts, and no identity mixing.

Figures

Figures reproduced from arXiv: 2604.14541 by Hao Pan, Hao Zhu, Jiahao Li, Jiawei Zhang, Lei Chu, Liqiang Liu, Yan Lu, Yanwen Wang, Yicheng Gong.

Figure 1
Figure 1. Figure 1: Emotion-Interpolated 3D Head Avatars. Given a single image, our feed-forward framework reconstructs expressive 3D avatars with explicit and interpolatable emotion control. The same blended emotion exhibits identity-aware variations, and all results are produced in a single forward pass without per-identity optimization. Abstract. We present a framework for explicit emotion control in feed￾forward, single-i… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the proposed framework. The system consists of (left) emotion￾consistent data curation and (right) dual-path emotion-aware modulation. We build a time-synchronized multi-identity dataset by transferring frame-aligned emotional dy￾namics from anchor subjects, yielding explicit emotion supervision disentangled from speech and identity. The reconstruction network modulates geometry and appearance … view at source ↗
Figure 3
Figure 3. Figure 3: Reconstruction and reenactment on a held-out synthetic dataset. Top: self￾identity reenactment using the subject’s own motion. Bottom: cross-identity reenact￾ment driven by another subject. Compared to the original baseline, emotion-aware modulation introduces no degradation in generation quality, consistently maintaining the same level of reconstruction fidelity and driving accuracy across settings and ba… view at source ↗
Figure 4
Figure 4. Figure 4: Emotion transfer with explicit emotion control. Rows show four identities (top: synthetic with GT; others: real), and columns compare methods. The target emotion is indicated on the left. Our approach enforces the specified emotion independent of source appearance or motion, yielding identity-consistent results [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Ablation of dual-path emotion modulation. Without appearance modulation, texture-level emotion leaks from the reference; without geometry modulation, deforma￾tion follows the driving motion. The full model suppresses both and matches the target emotion. Additionally, geometry reuse across backbones achieves comparable results, highlighting its robust transferability [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Emotion control across identities under fixed geometry. Rows: identities; columns: seven target emotions. All results share the same driving FLAME sequence. Identities respond differently to the same emotion, while each identity exhibits consis￾tent emotion-specific variation. content, and apply different emotion controls through the geometry modulation branch. As illustrated in [PITH_FULL_IMAGE:figures/f… view at source ↗
Figure 7
Figure 7. Figure 7: Motion robustness under fixed emotion. Rows fix identity and emotion; columns vary driving FLAME motion. Emotion remains consistent across articulations while preserving identity [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Continuous emotion interpolation. For a fixed identity, rows interpolate between happy–neutral–sad and disgust–fear–angry. Despite discrete training labels, embedding interpolation yields smooth geometry and appearance transitions. FLAME geometry is shown in the inset. sition in affect. Throughout the interpolation, identity-specific facial character￾istics remain stable, speech-driven articulation is pres… view at source ↗
read the original abstract

We present a framework for explicit emotion control in feed-forward, single-image 3D head avatar reconstruction. Unlike existing pipelines where emotion is implicitly entangled with geometry or appearance, we treat emotion as a first-class control signal that can be manipulated independently and consistently across identities. Our method injects emotion into existing feed-forward architectures via a dual-path modulation mechanism without modifying their core design. Geometry modulation performs emotion-conditioned normalization in the original parametric space, disentangling emotional state from speech-driven articulation, while appearance modulation captures identity-aware, emotion-dependent visual cues beyond geometry. To enable learning under this setting, we construct a time-synchronized, emotion-consistent multi-identity dataset by transferring aligned emotional dynamics across identities. Integrated into multiple state-of-the-art backbones, our framework preserves reconstruction and reenactment fidelity while enabling controllable emotion transfer, disentangled manipulation, and smooth emotion interpolation, advancing expressive and scalable 3D head avatars.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper presents a framework for explicit emotion control in feed-forward single-image 3D head avatar reconstruction. It treats emotion as a first-class independent control signal and injects it into existing architectures via a dual-path modulation mechanism (geometry modulation through emotion-conditioned normalization in parametric space to disentangle from speech-driven articulation, plus appearance modulation for identity-aware emotion cues) without altering the core backbone design. A key enabler is the construction of a time-synchronized, emotion-consistent multi-identity dataset via transfer of aligned emotional dynamics across identities. The method is claimed to preserve reconstruction/reenactment fidelity while supporting controllable emotion transfer, disentangled manipulation, and smooth interpolation.

Significance. If the central claims hold after validation, the work would be a meaningful contribution to 3D head avatar research by enabling practical, backbone-agnostic addition of explicit emotion control. The dual-path design and cross-identity dataset construction address a real entanglement issue in current pipelines. Credit is due for the emphasis on integration without core modifications and the focus on consistency across identities, which could improve scalability of expressive avatars.

major comments (2)
  1. [Abstract / Dataset Construction] Dataset construction (as described in the abstract): the central claim that emotion can be manipulated independently and consistently across identities rests on the transferred multi-identity corpus preserving disentanglement. No quantitative validation is supplied (e.g., identity classification accuracy on neutral frames, emotion consistency scores across transferred sequences, or metrics for misalignment/artifacts), which directly risks the dual-path modulation learning correlated rather than independent factors.
  2. [Abstract / Evaluation] Evaluation and results sections: the abstract asserts that the framework 'preserves reconstruction and reenactment fidelity' and enables 'controllable emotion transfer' when integrated into multiple state-of-the-art backbones, yet no quantitative results, ablation studies, baseline comparisons, or implementation details (losses, training procedure, modulation equations) are provided. This makes it impossible to assess whether the geometry normalization truly disentangles emotional state from articulation or introduces artifacts.
minor comments (1)
  1. [Abstract] The abstract uses the phrase 'giving faces their feelings back' in the title but does not clarify how this relates to prior implicit emotion handling in the literature; a brief positioning sentence would improve clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback and for recognizing the potential contribution of the dual-path modulation approach and the cross-identity dataset construction. We address each major comment below, providing clarifications and committing to revisions where the manuscript can be strengthened without misrepresenting the presented work.

read point-by-point responses

  1. Referee: [Abstract / Dataset Construction] Dataset construction (as described in the abstract): the central claim that emotion can be manipulated independently and consistently across identities rests on the transferred multi-identity corpus preserving disentanglement. No quantitative validation is supplied (e.g., identity classification accuracy on neutral frames, emotion consistency scores across transferred sequences, or metrics for misalignment/artifacts), which directly risks the dual-path modulation learning correlated rather than independent factors.

    Authors: We agree that explicit quantitative validation of the transferred dataset would better support the disentanglement claim. The manuscript describes the construction via transfer of aligned emotional dynamics to maintain time-synchronization and emotion consistency across identities, but does not report the suggested metrics. In the revised version we will add quantitative evaluations, including identity classification accuracy on neutral frames, emotion consistency scores across transferred sequences, and misalignment metrics, to demonstrate that the corpus preserves independent factors. revision: yes

  2. Referee: [Abstract / Evaluation] Evaluation and results sections: the abstract asserts that the framework 'preserves reconstruction and reenactment fidelity' and enables 'controllable emotion transfer' when integrated into multiple state-of-the-art backbones, yet no quantitative results, ablation studies, baseline comparisons, or implementation details (losses, training procedure, modulation equations) are provided. This makes it impossible to assess whether the geometry normalization truly disentangles emotional state from articulation or introduces artifacts.

    Authors: The abstract summarizes the claims at a high level. The full manuscript presents qualitative results across multiple backbones showing fidelity preservation and controllable transfer, along with the dual-path design rationale. However, we acknowledge that additional quantitative support would allow a more rigorous assessment of disentanglement. In the revision we will expand the evaluation section with quantitative metrics (e.g., reconstruction error, reenactment fidelity scores), ablation studies on each modulation path, baseline comparisons, and explicit details on losses, training procedure, and modulation equations. revision: yes

Circularity Check

0 steps flagged

No significant circularity; extends external backbones with additive modulation

full rationale

The paper's core derivation introduces a dual-path modulation (geometry normalization in parametric space plus appearance modulation) into existing feed-forward single-image 3D head avatar architectures without altering their core design. Dataset construction via cross-identity emotion transfer is presented as an enabling preprocessing step rather than a fitted or self-derived quantity. No equations, predictions, or uniqueness claims reduce to self-definition, fitted inputs renamed as outputs, or load-bearing self-citations. The framework is explicitly integrated into multiple state-of-the-art external backbones, preserving fidelity while adding controllability, rendering the chain self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the separability of emotion from geometry and appearance plus the feasibility of cross-identity emotion transfer; no free parameters or invented entities are explicitly introduced in the abstract.

axioms (1)
  • domain assumption Emotion can be treated as an independent control signal separable from identity, speech articulation, and appearance in 3D head models
    Invoked as the basis for the dual-path modulation mechanism and dataset construction.

pith-pipeline@v0.9.0 · 5483 in / 1212 out tokens · 28207 ms · 2026-05-10T11:45:10.959350+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

103 extracted references · 3 canonical work pages

  1. [1]

    In: CVPR

    Abdal, R., Lee, H.Y., Zhu, P., Chai, M., Siarohin, A., Wonka, P., Tulyakov, S.: 3davatargan: Bridging domains for personalized editable avatars. In: CVPR. pp. 4552--4562 (June 2023)

    2023

  2. [2]

    In: CVPR

    Abdal, R., Yifan, W., Shi, Z., Xu, Y., Po, R., Kuang, Z., Chen, Q., Yeung, D.Y., Wetzstein, G.: Gaussian shell maps for efficient 3d human generation. In: CVPR. pp. 9441--9451 (June 2024)

    2024

  3. [3]

    In: ICCV (2025)

    Aneja, S., Sevastopolsky, A., Kirschstein, T., Thies, J., Dai, A., Nie ner, M.: Gaussianspeech: Audio-driven gaussian avatars. In: ICCV (2025)

    2025

  4. [4]

    In: WACV

    Bhattarai, A.R., Nie ner, M., Sevastopolsky, A.: Triplanenet: An encoder for eg3d inversion. In: WACV. pp. 3055--3065 (2024)

    2024

  5. [5]

    In: SIGGRAPH

    Blanz, V., Vetter, T.: A morphable model for the synthesis of 3d faces. In: SIGGRAPH. pp. 187--194. ACM Press (1999)

    1999

  6. [6]

    In: CVPR (2021)

    Buehler, M.C., Meka, A., Li, G., Beeler, T., Hilliges, O.: Varitex: Variational neural face textures. In: CVPR (2021)

    2021

  7. [7]

    In: ICCV

    B \"u hler, M.C., Sarkar, K., Shah, T., Li, G., Wang, D., Helminger, L., Orts-Escolano, S., Lagun, D., Hilliges, O., Beeler, T., et al.: Preface: A data-driven volumetric prior for few-shot ultra high-resolution face synthesis. In: ICCV. pp. 3402--3413 (2023)

    2023

  8. [8]

    ACM TOG 41(4) (Jul 2022)

    Cao, C., Simon, T., Kim, J.K., Schwartz, G., Zollhoefer, M., Saito, S.S., Lombardi, S., Wei, S.E., Belko, D., Yu, S.I., Sheikh, Y., Saragih, J.: Authentic volumetric avatars from a phone scan. ACM TOG 41(4) (Jul 2022)

    2022

  9. [9]

    IEEE TVCG 20(3), 413--425 (2014)

    Cao, C., Weng, Y., Zhou, S., Tong, Y., Zhou, K.: Facewarehouse: A 3d facial expression database for visual computing. IEEE TVCG 20(3), 413--425 (2014)

    2014

  10. [10]

    In: CVPR (2021)

    Chan, E., Monteiro, M., Kellnhofer, P., Wu, J., Wetzstein, G.: pi-gan: Periodic implicit generative adversarial networks for 3d-aware image synthesis. In: CVPR (2021)

    2021

  11. [11]

    In: CVPR (2022)

    Chan, E.R., Lin, C.Z., Chan, M.A., Nagano, K., Pan, B., Mello, S.D., Gallo, O., Guibas, L., Tremblay, J., Khamis, S., Karras, T., Wetzstein, G.: Efficient geometry-aware 3D generative adversarial networks. In: CVPR (2022)

    2022

  12. [12]

    In: SIGGRAPH Conference Papers

    Chen, Y., Wang, L., Li, Q., Xiao, H., Zhang, S., Yao, H., Liu, Y.: Monogaussianavatar: Monocular gaussian point-based head avatar. In: SIGGRAPH Conference Papers. pp. 1--9 (2024)

    2024

  13. [13]

    In: NeurIPS

    Chu, X., Harada, T.: Generalizable and animatable gaussian head avatar. In: NeurIPS. vol. 37, pp. 57642--57670 (2024)

    2024

  14. [14]

    In: ICLR (2024)

    Chu, X., Li, Y., Zeng, A., Yang, T., Lin, L., Liu, Y., Harada, T.: Gpavatar: Generalizable and precise head avatar from image (s). In: ICLR (2024)

    2024

  15. [15]

    In: SIGGRAPH Asia Conference Papers

    Cui, J., Chen, Y., Xu, M., Shang, H., Chen, Y., Zhan, Y., Dong, Z., Yao, Y., Wang, J., Zhu, S.: Hallo4: High-fidelity dynamic portrait animation via direct preference optimization and temporal motion modulation. In: SIGGRAPH Asia Conference Papers. ACM (2025)

    2025

  16. [16]

    In: ICLR (2025)

    Cui, J., Li, H., Yao, Y., Zhu, H., Shang, H., Cheng, K., Zhou, H., Zhu, S., Wang, J.: Hallo2: Long-duration and high-resolution audio-driven portrait image animation. In: ICLR (2025)

    2025

  17. [17]

    In: CVPR (2025)

    Cui, J., Li, H., Zhan, Y., Shang, H., Cheng, K., Ma, Y., Mu, S., Zhou, H., Wang, J., Zhu, S.: Hallo3: Highly dynamic and realistic portrait image animation with video diffusion transformer. In: CVPR (2025)

    2025

  18. [18]

    In: CVPR

    Deng, Y., Wang, D., Ren, X., Chen, X., Wang, B.: Portrait4d: Learning one-shot 4d head avatar synthesis using synthetic data. In: CVPR. pp. 7119--7130 (2024)

    2024

  19. [19]

    In: ECCV (2024)

    Deng, Y., Wang, D., Wang, B.: Portrait4d-v2: Pseudo multi-view data creates better 4d head synthesizer. In: ECCV (2024)

    2024

  20. [20]

    In: ECCV (2024)

    Dhamo, H., Nie, Y., Moreau, A., Song, J., Shaw, R., Zhou, Y., Pérez-Pellitero, E.: Headgas: Real-time animatable head avatars via 3d gaussian splatting. In: ECCV (2024)

    2024

  21. [21]

    In: CVPR (2022)

    Fan, Y., Lin, Z., Saito, J., Wang, W., Komura, T.: Faceformer: Speech-driven 3d facial animation with transformers. In: CVPR (2022)

    2022

  22. [22]

    In: SIGGRAPH Asia Conference Papers (2024)

    Gao, X., Xiao, H., Zhong, C., Hu, S., Guo, Y., Zhang, J.: Portrait video editing empowered by multimodal generative priors. In: SIGGRAPH Asia Conference Papers (2024)

    2024

  23. [23]

    ACM TOG 41(6) (2022)

    Gao, X., Zhong, C., Xiang, J., Hong, Y., Guo, Y., Zhang, J.: Reconstructing personalized semantic facial nerf models from monocular video. ACM TOG 41(6) (2022)

    2022

  24. [24]

    In: SIGGRAPH Asia Conference Papers (2025)

    Gao, X., Zhou, J., Liu, D., Zhou, Y., Zhang, J.: Constructing diffusion avatar with learnable embeddings. In: SIGGRAPH Asia Conference Papers (2025)

    2025

  25. [25]

    In: SIGGRAPH Asia Conference Papers (2024)

    Giebenhain, S., Kirschstein, T., R \" u nz, M., Agapito, L., Nie ner, M.: Npga: Neural parametric gaussian avatars. In: SIGGRAPH Asia Conference Papers (2024)

    2024

  26. [26]

    In: CVPR (2025)

    Gu, Y., Tran, P., Zheng, Y., Xu, H., Li, H., Karmanov, A., Li, H.: Diffportrait360: Consistent portrait diffusion for 360 view synthesis. In: CVPR (2025)

    2025

  27. [27]

    In: ICMI (2023)

    Haque, K.I., Yumak, Z.: Facexhubert: Text-less speech-driven e (x) pressive 3d facial animation synthesis using self-supervised speech representation learning. In: ICMI (2023)

    2023

  28. [28]

    ACM TOG 44(4), 1--12 (2025)

    He, C., Li, J., Kirschstein, T., Sevastopolsky, A., Saito, S., Tan, Q., Romero, J., Cao, C., Rushmeier, H., Nam, G.: 3dgh: 3d head generation with composable hair and face. ACM TOG 44(4), 1--12 (2025)

    2025

  29. [29]

    In: ECCV

    He, Q., Ji, X., Gong, Y., Lu, Y., Diao, Z., Huang, L., Yao, Y., Zhu, S., Ma, Z., Xu, S., et al.: Emotalk3d: High-fidelity free-view synthesis of emotional 3d talking head. In: ECCV. Springer (2024)

    2024

  30. [30]

    In: SIGGRAPH Conference Papers

    He, Y., Gu, X., Ye, X., Xu, C., Zhao, Z., Dong, Y., Yuan, W., Dong, Z., Bo, L.: Lam: Large avatar model for one-shot animatable gaussian head. In: SIGGRAPH Conference Papers. pp. 1--13 (2025)

    2025

  31. [31]

    In: CVPR

    Hong, F.T., Zhang, L., Shen, L., Xu, D.: Depth-aware generative adversarial network for talking head video generation. In: CVPR. pp. 3397--3406 (2022)

    2022

  32. [32]

    In: CVPR

    Hong, Y., Peng, B., Xiao, H., Liu, L., Zhang, J.: Headnerf: A real-time nerf-based parametric head model. In: CVPR. pp. 20374--20384 (2022)

    2022

  33. [33]

    In: ICLR (2026)

    Ji, X., Weiss, S., Kansy, M., Naruniec, J., Cao, X., Solenthaler, B., Bradley, D.: Fast GHA : Generalized few-shot 3d gaussian head avatars with real-time animation. In: ICLR (2026)

    2026

  34. [34]

    In: SIGGRAPH Conference Papers

    Ji, X., Zhou, H., Wang, K., Wu, Q., Wu, W., Xu, F., Cao, X.: Eamm: One-shot emotional talking face via audio-based emotion-aware motion model. In: SIGGRAPH Conference Papers. SIGGRAPH '22 (2022)

    2022

  35. [35]

    In: CVPR

    Ji, X., Zhou, H., Wang, K., Wu, W., Loy, C.C., Cao, X., Xu, F.: Audio-driven emotional video portraits. In: CVPR. pp. 15480--15489 (2021)

    2021

  36. [36]

    ACM TOG 42(4), 139--1 (2023)

    Kerbl, B., Kopanas, G., Leimk \"u hler, T., Drettakis, G.: 3d gaussian splatting for real-time radiance field rendering. ACM TOG 42(4), 139--1 (2023)

    2023

  37. [37]

    In: CVPR (2026)

    Kirschstein, T., Giebenhain, S., Nie ner, M.: Flexavatar: Learning complete 3d head avatars with partial supervision. In: CVPR (2026)

    2026

  38. [38]

    In: SIGGRAPH Asia Conference Papers

    Kirschstein, T., Giebenhain, S., Tang, J., Georgopoulos, M., Nie ner, M.: GGHead: Fast and Generalizable 3D Gaussian Heads . In: SIGGRAPH Asia Conference Papers. SA '24, Association for Computing Machinery, New York, NY, USA (2024)

    2024

  39. [39]

    ACM TOG 42(4) (jul 2023)

    Kirschstein, T., Qian, S., Giebenhain, S., Walter, T., Nie ner, M.: Nersemble: Multi-view radiance field reconstruction of human heads. ACM TOG 42(4) (jul 2023)

    2023

  40. [40]

    In: ICCV (2025)

    Kirschstein, T., Romero, J., Sevastopolsky, A., Nie ner, M., Saito, S.: Avat3r: Large animatable gaussian reconstruction model for high-fidelity 3d head avatars. In: ICCV (2025)

    2025

  41. [41]

    In: ECCV (2024)

    Li, H., Chen, C., Shi, T., Qiu, Y., An, S., Chen, G., Han, X.: Spherehead: Stable 3d full-head synthesis with spherical tri-plane representation. In: ECCV (2024)

    2024

  42. [42]

    In: NeurIPS (2025)

    Li, H., Liu, K., Qiu, L., Zuo, Q., Zheng, K., Dong, Z., Han, X.: Hyplanehead: Rethinking tri-plane-like representations in full-head image synthesis. In: NeurIPS (2025)

    2025

  43. [43]

    In: ICLR (2026)

    Li, H., Zhang, H., Qiu, Y., Sun, Z., Zheng, K., Qiu, L., Li, P., Zuo, Q., Chen, C., Zheng, Y., et al.: Condition matters in full-head 3d gans. In: ICLR (2026)

    2026

  44. [44]

    In: CVPR

    Li, L., Li, Y., Weng, Y., Zheng, Y., Zhou, K.: Rgbavatar: Reduced gaussian blendshapes for online modeling of head avatars. In: CVPR. pp. 10747--10757 (June 2025)

    2025

  45. [45]

    ACM TOG 36(6), 194--1 (2017)

    Li, T., Bolkart, T., Black, M.J., Li, H., Romero, J.: Learning a model of facial shape and expression from 4d scans. ACM TOG 36(6), 194--1 (2017)

    2017

  46. [46]

    In: CVPR

    Li, W., Zhang, L., Wang, D., Zhao, B., Wang, Z., Chen, M., Zhang, B., Wang, Z., Bo, L., Li, X.: One-shot high-fidelity talking-head synthesis with deformable neural radiance field. In: CVPR. pp. 17969--17978 (2023)

    2023

  47. [47]

    In: ECCV

    Li, X., Cheng, Y., Ren, X., Jia, H., Xu, D., Zhu, W., Yan, Y.: Topo4d: Topology-preserving gaussian splatting for high-fidelity 4d head capture. In: ECCV. pp. 128--145. Springer (2024)

    2024

  48. [48]

    In: CVPR

    Li, X., Wang, J., Cheng, Y., Zeng, Y., Ren, X., Zhu, W., Zhao, W., Yan, Y.: Towards high-fidelity 3d talking avatar with personalized dynamic texture. In: CVPR. pp. 204--214 (2025)

    2025

  49. [49]

    In: NeurIPS

    Li, X., De Mello, S., Liu, S., Nagano, K., Iqbal, U., Kautz, J.: Generalizable one-shot 3d neural head avatar. In: NeurIPS. vol. 36 (2024)

    2024

  50. [50]

    In: CVPR (2026)

    Li, Z., Pun, C.M., Fang, C., Wang, J., Cun, X.: Personalive! expressive portrait image animation for live streaming. In: CVPR (2026)

    2026

  51. [51]

    In: CVPR (2026)

    Liu, C., Jing, T., Ma, C., Zhou, X., Lian, Z., Jin, Q., Yuan, H., Huang, S.S.: Emodifftalk: Emotion-aware diffusion for editable 3d gaussian talking head. In: CVPR (2026)

    2026

  52. [52]

    In: CVPR (2025)

    Liu, H., Wang, X., Wan, Z., Ma, Y., Chen, J., Fan, Y., Shen, Y., Song, Y., Chen, Q.: Avatarartist: Open-domain 4d avatarization. In: CVPR (2025)

    2025

  53. [53]

    In: SIGGRAPH Conference Papers

    Liu, H., Wang, X., Wan, Z., Shen, Y., Song, Y., Liao, J., Chen, Q.: Headartist: Text-conditioned 3d head generation with self score distillation. In: SIGGRAPH Conference Papers. SIGGRAPH '24, Association for Computing Machinery, New York, NY, USA (2024)

    2024

  54. [54]

    Ma, C., Tan, S., Pan, Y., Yang, J., Tong, X.: Esgaussianface: Emotional and stylized audio-driven facial animation via 3d gaussian splatting. TVCG pp. 1--12 (2026)

    2026

  55. [55]

    In: CVPR

    Ma, Z., Zhu, X., Qi, G.J., Lei, Z., Zhang, L.: Otavatar: One-shot talking face avatar with controllable tri-plane rendering. In: CVPR. pp. 16901--16910 (2023)

    2023

  56. [56]

    In: NeurIPS Track on Datasets and Benchmarks (2024)

    Martinez, J., Kim, E., Romero, J., Bagautdinov, T., Saito, S., Yu, S.I., Anderson, S., Zollhöfer, M., Team, C.A.: Codec Avatar Studio: Paired Human Captures for Complete, Driveable, and Generalizable Avatars . In: NeurIPS Track on Datasets and Benchmarks (2024)

    2024

  57. [57]

    In: ECCV (2020)

    Mildenhall, B., Srinivasan, P.P., Tancik, M., Barron, J.T., Ramamoorthi, R., Ng, R.: Nerf: Representing scenes as neural radiance fields for view synthesis. In: ECCV (2020)

    2020

  58. [58]

    arXiv preprint arXiv:2312.06400 (2023)

    Mir, A., Alonso, E., Mondragón, E.: Dit-head: High-resolution talking head synthesis using diffusion transformers. arXiv preprint arXiv:2312.06400 (2023)

    work page arXiv 2023

  59. [59]

    in-the-wild

    Paraperas Papantoniou, F., Filntisis, P.P., Maragos, P., Roussos, A.: Neural emotion director: Speech-preserving semantic control of facial expressions in "in-the-wild" videos. In: CVPR (2022)

    2022

  60. [60]

    In: ACM MM

    Prajwal, K.R., Mukhopadhyay, R., Namboodiri, V.P., Jawahar, C.: A lip sync expert is all you need for speech to lip generation in the wild. In: ACM MM. p. 484–492. MM '20, Association for Computing Machinery, New York, NY, USA (2020)

    2020

  61. [61]

    In: CVPR

    Qian, S., Kirschstein, T., Schoneveld, L., Davoli, D., Giebenhain, S., Nie ner, M.: Gaussianavatars: Photorealistic head avatars with rigged 3d gaussians. In: CVPR. pp. 20299--20309 (2024)

    2024

  62. [62]

    In: ICCV

    Richard, A., Zollh\"ofer, M., Wen, Y., de la Torre, F., Sheikh, Y.: Meshtalk: 3d face animation from speech using cross-modality disentanglement. In: ICCV. pp. 1173--1182 (October 2021)

    2021

  63. [63]

    In: ACCV

    Shen, X., Khan, F.F., Elhoseiny, M.: Emotalker: Audio driven emotion aware talking head generation. In: ACCV. pp. 1900--1917 (December 2024)

    1900

  64. [64]

    In: SIGGRAPH Conference Papers

    Song, L., Zhou, Y., Xu, Z., Zhou, Y., Aneja, D., Xu, C.: Streamme: Simplify 3d gaussian avatar within live stream. In: SIGGRAPH Conference Papers. SIGGRAPH '25, Association for Computing Machinery, New York, NY, USA (2025)

    2025

  65. [65]

    In: WACV

    Stypu kowski, M., Vougioukas, K., He, S., Zi e ba, M., Petridis, S., Pantic, M.: Diffused heads: Diffusion models beat gans on talking-face generation. In: WACV. pp. 5091--5100 (2024)

    2024

  66. [66]

    In: CVPR (2023)

    Sun, J., Wang, X., Wang, L., Li, X., Zhang, Y., Zhang, H., Liu, Y.: Next3d: Generative neural texture rasterization for 3d-aware head avatars. In: CVPR (2023)

    2023

  67. [67]

    ACM TOG 43(4) (2024)

    Sun, Z., Lv, T., Ye, S., Lin, M., Sheng, J., Wen, Y.H., Yu, M., Liu, Y.J.: Diffposetalk: Speech-driven stylistic 3d facial animation and head pose generation via diffusion models. ACM TOG 43(4) (2024)

    2024

  68. [68]

    In: ECCV

    Tan, S., Ji, B., Bi, M., Pan, Y.: Edtalk: Efficient disentanglement for emotional talking head synthesis. In: ECCV. pp. 398--416. Springer (2025)

    2025

  69. [69]

    In: ICCV

    Tan, S., Ji, B., Pan, Y.: Emmn: Emotional motion memory network for audio-driven emotional talking face generation. In: ICCV. pp. 22146--22156 (2023)

    2023

  70. [70]

    In: SIGGRAPH Asia Conference Papers

    Taubner, F., Zhang, R., Tuli, M., Bahmani, S., Lindell, D.B.: MVP4D : Multi-view portrait video diffusion for animatable 4D avatars. In: SIGGRAPH Asia Conference Papers. ACM (2025)

    2025

  71. [71]

    In: CVPR

    Taubner, F., Zhang, R., Tuli, M., Lindell, D.B.: CAP4D : Creating animatable 4D portrait avatars with morphable multi-view diffusion models. In: CVPR. pp. 5318--5330 (June 2025)

    2025

  72. [72]

    In: SIGGRAPH 2022 Conference Papers

    Wang, D., Chandran, P., Zoss, G., Bradley, D., Gotardo, P.: Morf: Morphable radiance fields for multiview neural head modeling. In: SIGGRAPH 2022 Conference Papers. SIGGRAPH '22, Association for Computing Machinery, New York, NY, USA (2022)

    2022

  73. [73]

    In: CVPR

    Wang, D., Deng, Y., Yin, Z., Shum, H.Y., Wang, B.: Progressive disentangled representation learning for fine-grained controllable talking head synthesis. In: CVPR. pp. 17979--17989 (2023)

    2023

  74. [74]

    TVCG (2025)

    Wang, J., Xie, J.C., Li, X., Xu, F., Pun, C.M., Gao, H.: Gaussianhead: High-fidelity head avatars with learnable gaussian derivation. TVCG (2025)

    2025

  75. [75]

    In: ECCV

    Wang, K., Wu, Q., Song, L., Yang, Z., Wu, W., Qian, C., He, R., Qiao, Y., Loy, C.C.: Mead: A large-scale audio-visual dataset for emotional talking-face generation. In: ECCV. Springer (2020)

    2020

  76. [76]

    In: CVPR (June 2022)

    Wang, L., Chen, Z., Yu, T., Ma, C., Li, L., Liu, Y.: Faceverse: a fine-grained and detail-controllable 3d face morphable model from a hybrid dataset. In: CVPR (June 2022)

    2022

  77. [77]

    In: CVPR

    Wang, T.C., Mallya, A., Liu, M.Y.: One-shot free-view neural talking-head synthesis for video conferencing. In: CVPR. pp. 10039--10049 (2021)

    2021

  78. [78]

    Aniportrait: Audio-driven synthesis of photorealistic portrait animation

    Wei, H., Yang, Z., Wang, Z.: Aniportrait: Audio-driven synthesis of photorealistic portrait animations. arXiv:2403.17694 (2024)

    work page arXiv 2024

  79. [79]

    In: ICLR (2025)

    Wu, T.W., Yang, J., Guo, Z., Wan, J., Zhong, F., Oztireli, C.: Gaussian head & shoulders: High fidelity neural upper body avatars with anchor gaussian guided texture warping. In: ICLR (2025)

    2025

  80. [80]

    In: CVPR (2026)

    Wu, Z., Zhou, B., Hu, L., Liu, H., Sun, Y., Wang, X., Cao, X., Shen, Y., Zhu, H.: Uika: Fast universal head avatar from pose-free images. In: CVPR (2026)

    2026

Showing first 80 references.