Pith. sign in

REVIEW 3 major objections 6 minor 63 references

Avatar-Forever: Decoupled Parallel Training for High-Quality Real-Time Infinite Avatars

T0 review · 3 major / 6 minor · reviewed 2026-08-16 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Decoupled training yields real-time infinite avatars at 27.2 FPS.

desk verdict Worth a serious referee, but the load-bearing decoupling claim rests on an additive merge that the paper does not actually isolate. read the letter →

arxiv 2608.12107 v1 pith:HWBPRF66 submitted 2026-08-12 cs.CV

Ruibin Li , Tao Yang , Zhiyuan Ma , Fangzhou Ai , Shilei Wen , Lei Zhang This is my paper · ORCID
classification cs.CV
keywords audio-drivenavatargenerationstreamingvideodistributionmatchingdistillationflowautoregressiverollouttrainingfeaturecachingsyntheticdatapipelinereal-timediffusioninference
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper argues that efficient few-step generation and long-horizon robustness are two independent capabilities of a streaming avatar generator, and that they should be trained in parallel rather than in a single sequential distillation pipeline. It builds a 22-billion-parameter video foundation model into a four-step generator by distillation in one branch, and trains a lightweight low-rank adapter in another branch to recover from errors that accumulate over autoregressive rollout. At inference the two updates are added to the original weights, producing one generator that is both fast and stable. A chunk-wise history feature cache removes redundant recomputation of fixed context during streaming. If the claims hold, avatar systems can be made real-time and unbounded in duration without the fragile, stage-coupled training pipelines currently used.

What carries the argument

The load-bearing mechanism is the additive composition of two separately trained parameter updates, implemented through Distribution Matching Distillation for the efficiency branch and Recovery-oriented Rollout Training with a masked flow-matching objective for the robustness branch. RRT is the conceptual engine: it simulates the inference-time failure mode by corrupting the earliest history chunk, letting the model generate several future chunks under its own predictions, and then supervising recovery only at the final target chunk, so the adapter learns to undo accumulated drift rather than one-step noise. ForeverCache is the inference-time mechanism: it performs a full-window forward pass at the first denoising step, caches per-block history features, and reuses them while denoising only the current chunk, removing redundant computation without changing the learned weights.

What would settle it

Run the trained system for several minutes on real conversational videos from a held-out dataset with camera noise, motion blur, and unseen lighting. If identity or lip-sync drift accumulates over time at the same rate as the distillation-only model, then the synthetic training distribution does not cover the failure modes that matter, and the central claim fails.

Watch

Extended reading notes

Core claim

The central claim is the composition identity $\theta^* = \theta_0 + \Delta\theta_{\mathrm{DMD}} + \Delta\theta_{\mathrm{RRT}}$: starting from the pretrained base weights $\theta_0$, a full-parameter distillation branch learns the few-step update $\Delta\theta_{\mathrm{DMD}}$, and a separate low-rank adaptation branch learns the long-horizon update $\Delta\theta_{\mathrm{RRT}}$; the two updates can be trained independently and then merged additively. The robustness branch uses Recovery-oriented Rollout Training (RRT), which degrades an early history chunk, rolls out several future chunks without gradients, and applies a standard flow-matching loss only to the next ground-truth chunk after the degradation has propagated. The efficiency branch never sees autoregressive rollout, and the robustness branch never distills. Together with the ForeverCache inference-time cache, the final generator is claimed to support unbounded audio-driven avatar generation at 768x512 resolution with end-to-end throughput of 27.2 FPS on a single H100 GPU, while maintaining identity consistency, motion coherence, and visual fidelity.

Load-bearing premise

The load-bearing premise is that videos synthesized by the same base video model, after automatic filtering, form a valid training distribution for long-horizon robustness, so the adapter trained on them transfers to real human videos with real lighting, noise, and recording conditions.

Editorial extensions

If this is right

  • Few-step efficiency and long-horizon robustness can be optimized by separate objectives and composed additively, so training pipelines become simpler to diagnose and scale.
  • Corrupted-history training alone is insufficient: RRT's benefit comes from supervising recovery after model-induced errors have propagated through several rollout chunks, with $K=4$ giving the most stable results.
  • ForeverCache raises streaming throughput by roughly 23% on short videos and 45% on 30-second generation while preserving most perceptual gains, with no retraining.
  • A fully synthetic data pipeline from the same base model can supply long-horizon avatar training data without large-scale curation of real long videos.
  • The combined system supports continuous generation beyond 11 minutes without visible drift, with 27.2 FPS end-to-end throughput at 768x512 on a single H100.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper's claims, the same decoupling principle suggests that any streaming video task with autoregressive drift could train a robustness adapter in parallel with a distillation branch and merge them additively.
  • Because the adapter is trained on videos produced by the same base model, its robustness may be bounded by error modes the base model can produce; mixing in real degraded footage or adversarially perturbed synthetic frames is a testable extension the paper does not explore.
  • ForeverCache's insight, caching stable context features across denoising steps, could extend to other fixed conditioning streams such as audio features or reference images, potentially cutting latency further.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper proposes Avatar-Forever, a framework for long-horizon audio-driven avatar generation built on a 22B video foundation model. It decouples training into an efficiency branch (full-parameter DMD distillation to a four-step generator) and a robustness branch (Recovery-oriented Rollout Training, RRT, a LoRA adapter trained by perturbing early history, rolling out K chunks without gradient, and applying flow-matching supervision only to the following target chunk). At deployment the two deltas are added to the base weights (Eq. 6), and ForeverCache caches historical chunk features after the first denoising step and forwards only current tokens in later steps. Training data are synthesized by the same base model from filtered dialogue prompts. Experiments on EMTD, HDTF, and TalkVid compare 5s and 30s clips against four prior systems using an LLM judge, automatic metrics, and a 20-person user study, reporting 27.2 FPS at 768x512 on an H100.

Significance. The significance is potentially high if the central claim holds: few-step efficiency and long-horizon robustness can be trained independently and composed additively, and history features can be cached chunk-wise without quality loss. The manuscript has several genuine strengths: the RRT objective is a simple, well-motivated standard flow-matching loss under autoregressive rollout; the paper publishes the evaluation prompt in an appendix; the decoupling hypothesis is tested with a relevant three-way ablation (Table 3); and the external evaluation spans three datasets. Falsifiable claims are clearly stated (Eq. 6, 27.2 FPS, 23% throughput gain). However, the load-bearing conditions—transferability of the RRT delta across denoising schedules, validity of self-synthetic training data for real-world robustness, and the exactness of ForeverCache—are not yet established, so I cannot recommend acceptance without revision.

major comments (3)
  1. [Section 3.2, Eq. (6); Section 5.1] The central claim that the RRT LoRA delta, trained against the 30-step base-model rollout, transfers to the four-step DMD generator is not isolated by any experiment. Table 3 shows that adding RRT to DMD improves over DMD-only, but it does not rule out that the improvement comes from a generic LoRA on the DMD student, nor does it test whether a correction calibrated under 30-step rollout errors is valid under the 4-step inference used at deployment. Please add a sequential control (RRT trained on the DMD student), a 4-step rollout training control, and a same-rank random-LoRA control; without these, Eq. (6) is an untested composition rule rather than evidence for decoupled parallel training.
  2. [Section 4 and Section 5.1] The training corpus is synthesized by the same LTX-2.3 model that Avatar-Forever adapts, so RRT is trained on the base model's own output distribution; failure modes that occur only with real speakers, natural lighting, camera noise, or recording artifacts are absent from training. The external evaluation uses 40-sample splits per dataset, the principal perceptual metric is an author-designed LLM judge without reported agreement with human ratings or run-to-run variance, Table 1 reports no confidence intervals, and the extended-duration claim rests on a single 11-minute qualitative example. Please report error bars and per-sample variance, validate the LLM judge against the human study (or report inter-rater agreement), and include a longer real-video evaluation or a targeted domain-shift test before claiming 'unbounded' real-world avatar generation.
  3. [Section 3.3, Eqs. (8)-(9)] ForeverCache's correctness rests on the assumption that feature representations of clean historical chunks computed at the first denoising step remain valid at later noise levels, but Eq. (8) and Eq. (9) simply define two execution modes and do not establish that the cached attention values approximate the full-window forward pass. The only evidence is the global gap between 'Ours' and 'Ours w/ FC' in Table 1 and qualitative figures; no per-step velocity error or a cached-vs-recomputed ablation on identical chunks is reported. A direct measurement of the approximation error is needed to support the claim that the cache preserves the denoising interface and removes redundant computation without quality loss.
minor comments (6)
  1. [Section 2.2] The phrase 'complicated foring strategies' appears to be a typo for 'complicated forcing strategies'.
  2. [Figure 1] The label '在此处键入公式' in Figure 1 is an untranslated placeholder and should be removed or replaced with an English caption.
  3. [Abstract and Section 5.2] The abstract's 'unbounded' claim is stronger than the evidence, which covers 5s and 30s splits plus one 11-minute qualitative example; please qualify the wording accordingly.
  4. [Section 3.1 and Section 5.1] The DMD reference is inconsistent: Section 3.1 cites [44] while Section 5.1 cites [43]; please use the correct reference for the DMD variant actually adopted.
  5. [Table 1] The OmniAvatar latency entry '850.00/>1h' uses inconsistent units and formatting; please define the notation and ensure all latency entries are comparable.
  6. [Section 3.2, Eq. (3)] In Eq. (3), the notation sg(G_theta(...)) is used for a single denoising step, but G_theta elsewhere denotes the full generator; please clarify the sampler-step notation.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: central claims are empirically tested on external benchmarks; Eq. (6) is a deployment heuristic, not a derived prediction.

full rationale

The paper's load-bearing claims—(i) decoupled DMD and RRT branches can be trained in parallel and merged, (ii) RRT improves long-horizon stability, and (iii) ForeverCache accelerates streaming without degrading quality—are supported by ablations (Table 3, Figs. 8 and 9) and by external real-video evaluation on EMTD, HDTF, and TalkVid, rather than by equations that define the conclusion into the premise. Equations (1) and (5) are standard DMD and flow-matching objectives, and Eq. (6), θ* = θ0 + Δθ_DMD + Δθ_RRT, is an explicit composition design for the two branches, not a theorem whose output is presupposed by its inputs. The fact that RRT is rolled out with a 30-step schedule and merged into a 4-step student (Sec. 5.1) creates an empirical transfer risk, which is a correctness concern, not a derivational circularity. The synthetic training corpus is generated by the same LTX base model being adapted (Sec. 4), giving a self-training regime and a generalization-risk concern; however, the evaluation on real external datasets is independent and could have falsified the transfer, and no fitted parameter is relabeled as a prediction. Self-citations [7], [48], and [50] are contextual background or related-work references; none is invoked as an authority to force the method's central choices. Therefore no circular step meeting the evidentiary bar is present.

Assumptions & free parameters 9 free parameters · 5 assumptions · 0 invented entities

The central empirical claim relies on the base model, the synthetic self-training data, the additive composition of two independently trained branches, the hand-designed degradation family, and an unanalyzed feature-cache approximation. These are pragmatic design choices, but each is load-bearing and none is derived from first principles.

free parameters (9)
  • RRT rollout horizon K = 4
    Selected by the ablation in Figure 9; K=4 gives the most stable long-horizon results, so it is tuned on the evaluation metric rather than derived.
  • History degradation probability p = 0.5
    Chosen by hand in Section 5.1; no sensitivity analysis is reported.
  • LoRA rank / alpha = 128 / 128
    Set in Section 5.1; common high-rank setting, not justified by experiments.
  • Learning rate = 1e-5
    AdamW, fixed for both branches; no schedule reported.
  • Training steps = DMD 5000, RRT 3000
    Stopping points chosen by convergence; not justified.
  • Context and target chunk size = 4 latent frames / 4 latent frames
    Section 5.1; no ablation on chunk size.
  • Rollout denoising steps T = 30
    Uses the base sampler default without CFG; a design choice for the rollout trajectory.
  • Distilled generator sampling steps = 4
    DMD compresses to a four-step generator; chosen as the efficiency target.
  • LLM judge weights = 0.35 A-V, 0.35 visual, 0.30 motion
    Defined in Appendix A to compute the overall score; a subjective evaluation design choice.
assumptions (5)
  • domain assumption The LTX-2.3 base model provides a sufficiently strong video prior for avatar generation.
    The entire method is built on this 22B model (reference [6]); its capabilities are taken as given.
  • ad hoc to paper Synthetic videos generated by the same base model are distributed closely enough to the deployment distribution that training on them transfers to real data.
    Section 4 constructs all training data from LTX itself; no real long-horizon avatar videos are used.
  • ad hoc to paper DMD and RRT updates can be composed additively (Eq. 6) without destructive interference.
    The final generator sums a dense distilled update and a LoRA adapter trained separately; no theoretical guarantee is given, only the Table 3 ablation.
  • domain assumption The degradation operator D in Eq. 2, with noise, blur, saturation, latent masking, and identity mapping, simulates the autoregressive error distribution encountered at inference.
    RRT's trigger for rollout drift is a hand-designed corruption family; its fidelity to real accumulated errors is not measured.
  • ad hoc to paper Feature reuse in ForeverCache (Eqs. 8-9) preserves the velocity prediction to within acceptable error.
    The paper assumes cached history features are stable but provides no error analysis; the approximation is untested in isolation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Avatar-Forever: Decoupled Parallel Training for High-Quality Real-Time Infinite Avatars." pith.science (2026). https://pith.science/paper/HWBPRF66

@misc{pith2026260812107,
  author       = {Pith},
  title        = {Pith review of: Avatar-Forever: Decoupled Parallel Training for High-Quality Real-Time Infinite Avatars},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HWBPRF66}},
  note         = {Machine review of arXiv:2608.12107}
}
read the original abstract

Existing streaming video systems often rely on sequential, distillation-centered training pipelines to enable few-step long-video generation. However, this paradigm suffers from two limitations. First, failures or distribution shifts introduced in earlier stages affect later optimization, complicating the training process to converge. Second, the distillation-centric objective favours short-term generation but is prone to quality degradation when autoregressive errors accumulate over long rollouts. We propose Avatar-Forever, a decoupled parallel training framework for high-quality real-time infinite interactive avatars. Instead of coupling generation efficiency and long-horizon robustness under a sequential distillation pipeline, we treat them as two independent capabilities that can be trained in parallel. One branch performs full-parameter distillation to train an efficient generator with high visual quality, while another trains a lightweight long-horizon adapter via Recovery-oriented Rollout Training (RRT), which improves generation robustness under long-horizon inference conditions. Our decoupled parallel training design simplifies the overall training process and avoids unnecessary objective conflicts between few-step generation and long-horizon adaptation. We further introduce ForeverCache, a chunk-wise feature caching mechanism to substantially reduce redundant history computation during streaming inference. Built upon a 22B video foundation model, Avatar-Forever supports unbounded audio-driven avatar generation while maintaining identity consistency, motion coherence, and visual fidelity, enabling an end-to-end throughput of high-resolution 768x512 videos at 27.2 FPS on a single H100 GPU and providing a practical path toward stable digital humans.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

63 extracted references · 23 canonical work pages

  1. [1]

    Sadtalker: Learningrealistic3dmotioncoefficientsforstylizedaudio-drivensingleimagetalkingfaceanimation

    Wenxuan Zhang, Xiaodong Cun, Xuan Wang, Yong Zhang, Xi Shen, Yu Guo, Ying Shan, and Fei Wang. Sadtalker: Learningrealistic3dmotioncoefficientsforstylizedaudio-drivensingleimagetalkingfaceanimation. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8652–8661, 2023

    work page 2023

  2. [2]

    Latentsync: Audio conditioned latent diffusion models for lip sync.arXiv e-prints, pages arXiv–2412, 2024

    Chunyu Li, Chao Zhang, Weikai Xu, Jinghui Xie, Weiguo Feng, Bingyue Peng, and Weiwei Xing. Latentsync: Audio conditioned latent diffusion models for lip sync.arXiv e-prints, pages arXiv–2412, 2024

    work page 2024

  3. [3]

    Emo: Emote portrait alive: Generating expressive portrait videos with audio2video diffusion model under weak conditions.arXiv preprint arXiv:2402.17485, 2024

    Linrui Tian, Qi Wang, Bang Zhang, and Liefeng Bo. Emo: Emote portrait alive: Generating expressive portrait videos with audio2video diffusion model under weak conditions.arXiv preprint arXiv:2402.17485, 2024. Visual Computing Lab·The Hong Kong Polytechnic University 14 / 24

    arXiv 2024

  4. [4]

    Seedance 2.0: Advancing video generation for world complexity.arXiv preprint arXiv:2604.14148, 2026

    Team Seedance. Seedance 2.0: Advancing video generation for world complexity.arXiv preprint arXiv:2604.14148, 2026

    arXiv 2026

  5. [5]

    Wan: Openandadvancedlarge-scalevideogenerativemodels.arXivpreprintarXiv:2503.20314, 2025

    Team Wan, Ang Wang, Baole Ai, Bin Wen, Chaojie Mao, Chen-Wei Xie, Di Chen, Feiwu Yu, Haiming Zhao, JianxiaoYang,etal. Wan: Openandadvancedlarge-scalevideogenerativemodels.arXivpreprintarXiv:2503.20314, 2025

    arXiv 2025

  6. [6]

    Ltx-2: Efficient joint audio-visual foundation model

    Yoav HaCohen, Benny Brazowski, Nisan Chiprut, Yaki Bitterman, Andrew Kvochko, Avishai Berkowitz, Daniel Shalem, Daphna Lifschitz, Dudu Moshe, Eitan Porat, et al. Ltx-2: Efficient joint audio-visual foundation model. arXiv preprint arXiv:2601.03233, 2026

    arXiv 2026

  7. [7]

    Many-for-many: Unify the training of multiple video and image generation and manipulation tasks.arXiv preprint arXiv:2506.01758, 2025

    Tao Yang, Ruibin Li, Yangming Shi, Yuqi Zhang, Qide Dong, Haoran Cheng, Weiguo Feng, Shilei Wen, Bingyue Peng, and Lei Zhang. Many-for-many: Unify the training of multiple video and image generation and manipulation tasks.arXiv preprint arXiv:2506.01758, 2025

    arXiv 2025

  8. [8]

    SII-GAIR, Sand. ai, Ethan Chern, Hansi Teng, Hanwen Sun, Hao Wang, Hong Pan, Hongyu Jia, Jiadi Su, Jin Li, Junjie Yu, Lijie Liu, Lingzhi Li, Lyumanshan Ye, Min Hu, Qiangang Wang, Quanwei Qi, Steffi Chern, Tao Bu, Taoran Wang, Teren Xu, Tianning Zhang, Tiantian Mi, Weixian Xu, Wenqiang Zhang, Wentai Zhang, Xianping Yi, Xiaojie Cai, Xiaoyang Kang, Yan Ma, Y...

    arXiv 2026

Show all 63 references
  1. [9]

    Arig: Autoregressive interactive head generation for real-time conversations

    Ying Guo, Xi Liu, Cheng Zhen, Pengfei Yan, and Xiaoming Wei. Arig: Autoregressive interactive head generation for real-time conversations. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 12956–12965, 2025

    2025

  2. [10]

    Ditto: Motion-space diffusion for controllable realtime talking head synthesis

    Tianqi Li, Ruobing Zheng, Minghui Yang, Jingdong Chen, and Ming Yang. Ditto: Motion-space diffusion for controllable realtime talking head synthesis. InProceedings of the 33rd ACM International Conference on Multimedia, pages 9704–9713, 2025

    2025

  3. [11]

    Talkingmachines: Real-time audio-driven facetime-style video via autoregressive diffusion models.arXiv preprint arXiv:2506.03099, 2025

    Chetwin Low and Weimin Wang. Talkingmachines: Real-time audio-driven facetime-style video via autoregressive diffusion models.arXiv preprint arXiv:2506.03099, 2025

    2025 arXiv

  4. [12]

    Avatarforcing: One-step streaming talking avatars via local-future sliding-window denoising.arXiv preprint arXiv:2603.14331, 2026

    Liyuan Cui, Wentao Hu, Wenyuan Zhang, Zesong Yang, Fan Shi, and Xiaoqiang Liu. Avatarforcing: One-step streaming talking avatars via local-future sliding-window denoising.arXiv preprint arXiv:2603.14331, 2026

    2026

  5. [13]

    From slow bidirectional to fast autoregressive video diffusion models

    Tianwei Yin, Qiang Zhang, Richard Zhang, William T Freeman, Fredo Durand, Eli Shechtman, and Xun Huang. From slow bidirectional to fast autoregressive video diffusion models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 22963–22974, 2025

    2025

  6. [14]

    Self forcing: Bridging the train-test gap in autoregressive video diffusion.Advances in Neural Information Processing Systems, 38:167283–167308, 2026

    Xun Huang, Zhengqi Li, Guande He, Mingyuan Zhou, and Eli Shechtman. Self forcing: Bridging the train-test gap in autoregressive video diffusion.Advances in Neural Information Processing Systems, 38:167283–167308, 2026

    2026

  7. [15]

    Streamavatar: Streaming diffusion models for real-time interactive human avatars.arXiv preprint arXiv:2512.22065, 2026

    Zhiyao Sun, Ziqiao Peng, Yifeng Ma, Yi Chen, Zhengguang Zhou, Zixiang Zhou, Guozhen Zhang, Youliang Zhang, Yuan Zhou, Qinglin Lu, and Yong-Jin Liu. Streamavatar: Streaming diffusion models for real-time interactive human avatars.arXiv preprint arXiv:2512.22065, 2026. Accepted ...

    2026

  8. [16]

    Longlive: Real-time interactive long video generation.arXiv preprint arXiv:2509.22622, 2025

    ShuaiYang,WeiHuang,RuihangChu,YichengXiao,YuyangZhao,XianbangWang,MuyangLi,EnzeXie,Yingcong Chen, Yao Lu, et al. Longlive: Real-time interactive long video generation.arXiv preprint arXiv:2509.22622, 2025

    2025 arXiv

  9. [17]

    Reward forcing: Efficient streaming video generation with rewarded distribution matching distillation

    Yunhong Lu, Yanhong Zeng, Haobo Li, Hao Ouyang, Qiuyu Wang, Ka Leong Cheng, Jiapeng Zhu, Hengyuan Cao, Zhipeng Zhang, Xing Zhu, et al. Reward forcing: Efficient streaming video generation with rewarded distribution matching distillation. InProceedings of the IEEE/CVF Conferenc...

    2026

  10. [18]

    Stable video infinity: Infinite-length video generation with error recycling.arXiv preprint arXiv:2510.09212, 2025

    Wuyang Li, Wentao Pan, Po-Chien Luan, Yang Gao, and Alexandre Alahi. Stable video infinity: Infinite-length video generation with error recycling.arXiv preprint arXiv:2510.09212, 2025

    2025

  11. [19]

    Causal forcing: Autoregressive diffu- sion distillation done right for high-quality real-time interactive video generation.arXiv preprint arXiv:2602.02214, 2026

    Hongzhou Zhu, Min Zhao, Guande He, Hang Su, Chongxuan Li, and Jun Zhu. Causal forcing: Autoregressive diffu- sion distillation done right for high-quality real-time interactive video generation.arXiv preprint arXiv:2602.02214, 2026

    2026 arXiv

  12. [20]

    Infinity-rope: Action-controllable infinite video generation emerges from autoregressive self-rollout

    Hidir Yesiltepe, Tuna Meral, Adil Kaan Akan, Kaan Oktay, and Pinar Yanardag. Infinity-rope: Action-controllable infinite video generation emerges from autoregressive self-rollout. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 40256–...

    2026

  13. [21]

    Helios: Real real-time long video generation model.arXiv preprint arXiv:2603.04379, 2026

    Shenghai Yuan, Yuanyang Yin, Zongjian Li, Xinwei Huang, Xiao Yang, and Li Yuan. Helios: Real real-time long video generation model.arXiv preprint arXiv:2603.04379, 2026

    2026

  14. [22]

    Longlive-2.0: An nvfp4 parallel infrastructure for long video generation.arXiv preprint arXiv:2605.18739, 2026

    Yukang Chen, Luozhou Wang, Wei Huang, Shuai Yang, Bohan Zhang, Yicheng Xiao, Ruihang Chu, Weian Mao, Qixin Hu, Shaoteng Liu, et al. Longlive-2.0: An nvfp4 parallel infrastructure for long video generation.arXiv preprint arXiv:2605.18739, 2026

    2026 arXiv

  15. [23]

    Lpm 1.0: Video-based character performance model.arXiv preprint arXiv:2604.07823, 2026

    Ailing Zeng, Casper Yang, Chauncey Ge, Eddie Zhang, Garvey Xu, Gavin Lin, Gilbert Gu, Jeremy Pi, Leo Li, Mingyi Shi, Shawn Wang, Sheng Bi, Steven Tang, Thorn Hang, Tobey Guo, Vincent Li, Xin Tong, Yikang Li, Yuchen Sun, Yue Zhao, Yuhan Lu, Yuwei Li, Zane Zhang, Zeshi Yang, and...

    2026 arXiv

  16. [24]

    Soulx-flashtalk: Real-time infinite streaming of audio-driven avatars via self-correcting bidirectional distillation.arXiv preprint arXiv:2512.23379, 2026

    LeShen,QianQiao,TanYu,KeZhou,TianhangYu,YuZhan,ZhenjieWang,DingchengZhen,MingTao,Shunshun Yin, and Siyuan Liu. Soulx-flashtalk: Real-time infinite streaming of audio-driven avatars via self-correcting bidirectional distillation.arXiv preprint arXiv:2512.23379, 2026

    2026

  17. [25]

    Self-forcing++: Towards minute-scale high-quality video generation.arXiv preprint arXiv:2510.02283, 2025

    Justin Cui, Jie Wu, Ming Li, Tao Yang, Xiaojie Li, Rui Wang, Andrew Bai, Yuanhao Ban, and Cho-Jui Hsieh. Self-forcing++: Towards minute-scale high-quality video generation.arXiv preprint arXiv:2510.02283, 2025

    2025 arXiv

  18. [26]

    Infinitetalk: Audio-driven video generation for sparse-frame video dubbing.arXiv preprint arXiv:2508.14033, 2025

    Shaoshu Yang, Zhe Kong, Feng Gao, Meng Cheng, Xiangyu Liu, Yong Zhang, Zhuoliang Kang, Wenhan Luo, Xunliang Cai, Ran He, et al. Infinitetalk: Audio-driven video generation for sparse-frame video dubbing.arXiv preprint arXiv:2508.14033, 2025

    2025 arXiv

  19. [27]

    Live avatar: Streaming real-time audio-driven avatar generation with infinite length.arXiv preprint arXiv:2512.04677, 2025

    Yubo Huang, Hailong Guo, Fangtai Wu, Weiqiang Wang, Shifeng Zhang, Shijie Huang, Qijun Gan, Lin Liu, Sirui Zhao, Enhong Chen, Jiaming Liu, and Steven Hoi. Live avatar: Streaming real-time audio-driven avatar generation with infinite length.arXiv preprint arXiv:2512.04677, 2025

    2025 arXiv

  20. [28]

    Style transfer for 2d talking head generation

    Trong Thang Pham, Tuong Do, Nhat Le, Ngan Le, Hung Nguyen, Erman Tjiputra, Quang Tran, and Anh Nguyen. Style transfer for 2d talking head generation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7500–7509, 2024

    2024

  21. [29]

    Disentangledandcontrollablefaceimagegeneration via 3d imitative-contrastive learning

    YuDeng,JiaolongYang,DongChen,FangWen,andXinTong. Disentangledandcontrollablefaceimagegeneration via 3d imitative-contrastive learning. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5154–5163, 2020

    2020

  22. [30]

    Capture, learning, and synthesis of 3d speaking styles

    Daniel Cudeiro, Timo Bolkart, Cassidy Laidlaw, Anurag Ranjan, and Michael J Black. Capture, learning, and synthesis of 3d speaking styles. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10101–10111, 2019

    2019

  23. [31]

    A lip sync expert is all you need for speech to lip generation in the wild

    K R Prajwal, Rudrabha Mukhopadhyay, Vinay Namboodiri, and C V Jawahar. A lip sync expert is all you need for speech to lip generation in the wild. InProceedings of the 28th ACM International Conference on Multimedia, pages 484–492, 2020

    2020

  24. [32]

    Dreamavatar: Text-and-shape guided 3d human avatar generation via diffusion models

    Yukang Cao, Yan-Pei Cao, Kai Han, Ying Shan, and Kwan-Yee K Wong. Dreamavatar: Text-and-shape guided 3d human avatar generation via diffusion models. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 958–968, 2024

    2024

  25. [33]

    Latentavatar: Learninglatentexpressioncodeforexpressiveneuralheadavatar

    Yuelang Xu, Hongwen Zhang, Lizhen Wang, Xiaochen Zhao, Han Huang, Guojun Qi, and Yebin Liu. Latentavatar: Learninglatentexpressioncodeforexpressiveneuralheadavatar. InACMSIGGRAPH2023ConferenceProceedings, pages 1–10, 2023

    2023

  26. [34]

    Denoisingdiffusionprobabilisticmodels.Advancesinneuralinformation processing systems, 33:6840–6851, 2020

    JonathanHo,AjayJain,andPieterAbbeel. Denoisingdiffusionprobabilisticmodels.Advancesinneuralinformation processing systems, 33:6840–6851, 2020

    2020

  27. [35]

    Flow matching for generative modeling.arXiv preprint arXiv:2210.02747, 2022

    Yaron Lipman, Ricky TQ Chen, Heli Ben-Hamu, Maximilian Nickel, and Matt Le. Flow matching for generative modeling.arXiv preprint arXiv:2210.02747, 2022

    2022 arXiv

  28. [36]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image synthesis with latent diffusion models. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022

    2022

  29. [37]

    Difftalk: Crafting diffusion models for generalized audio-driven portraits animation

    Shuai Shen, Wenliang Zhao, Zibin Meng, Wanhua Li, Zheng Zhu, Jie Zhou, and Jiwen Lu. Difftalk: Crafting diffusion models for generalized audio-driven portraits animation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1982–1991, 2023

    1982

  30. [38]

    Hallo: Hierarchical audio-driven visual synthesis for portrait image animation.arXiv preprint arXiv:2406.08801, 2024

    Mingwang Xu, Hui Li, Qingkun Su, Hanlin Shang, Liwei Zhang, Ce Liu, Jingdong Wang, Yao Yao, and Siyu Zhu. Hallo: Hierarchical audio-driven visual synthesis for portrait image animation.arXiv preprint arXiv:2406.08801, 2024. Visual Computing Lab·The Hong Kong Polytechnic Univer...

    2024 arXiv

  31. [39]

    Aniportrait: Audio-driven synthesis of photorealistic portrait animation.arXiv preprint arXiv:2403.17694, 2024

    Huawei Wei, Zejun Yang, and Zhisheng Wang. Aniportrait: Audio-driven synthesis of photorealistic portrait animation.arXiv preprint arXiv:2403.17694, 2024

    2024 arXiv

  32. [40]

    Echomimic: Lifelike audio-driven portrait animations through editable landmark conditioning.arXiv preprint arXiv:2407.08136, 2024

    Zhiyuan Chen, Jiajiong Cao, Zhiquan Chen, Yuming Li, and Chenguang Ma. Echomimic: Lifelike audio-driven portrait animations through editable landmark conditioning.arXiv preprint arXiv:2407.08136, 2024

    2024 arXiv

  33. [41]

    Kling-omni technical report.arXiv preprint arXiv:2512.16776, 2025

    Kling Team. Kling-omni technical report.arXiv preprint arXiv:2512.16776, 2025

    2025 arXiv

  34. [42]

    Video generation models as world simulators

    OpenAI. Video generation models as world simulators. https://openai.com/index/ video-generation-models-as-world-simulators/, 2024. Technical report

    2024

  35. [43]

    Freeman, and Taesung Park

    Tianwei Yin, Michaël Gharbi, Richard Zhang, Eli Shechtman, Frédo Durand, William T. Freeman, and Taesung Park. One-step diffusion with distribution matching distillation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6613–6623, 2024

    2024

  36. [44]

    Freeman, and Taesung Park

    Tianwei Yin, Micha"el Gharbi, Richard Zhang, Eli Shechtman, Frédo Durand, William T. Freeman, and Taesung Park. Improved distribution matching distillation for fast image synthesis.arXiv preprint arXiv:2405.14867, 2024

    2024 arXiv

  37. [45]

    Adversarial diffusion distillation

    Axel Sauer, Dominik Lorenz, Andreas Blattmann, Puneet Dokania, Stefano Ermon, Andreas Geiger, Patrick Esser, and Robin Rombach. Adversarial diffusion distillation. InEuropean Conference on Computer Vision, pages 87–103, 2024

    2024

  38. [46]

    Latent consistency models: Synthesizing high-resolution images with few-step inference.arXiv preprint arXiv:2310.04378, 2023

    Simian Luo, Yiqin Tan, Longbo Huang, Jian Li, and Hang Zhao. Latent consistency models: Synthesizing high-resolution images with few-step inference.arXiv preprint arXiv:2310.04378, 2023

    2023 arXiv

  39. [47]

    Phased consistency models

    Fu-Yun Wang, Zhaoyang Huang, Alexander William Bergman, Dazhong Shen, Peng Gao, Michael Lingelbach, Keqiang Sun, Weikang Bian, Guanglu Song, Yu Liu, Xiaogang Wang, and Hongsheng Li. Phased consistency models. InAdvances in Neural Information Processing Systems, 2024

    2024

  40. [48]

    Diversity-preserveddistributionmatchingdistillationforfastvisual synthesis.arXiv preprint arXiv:2602.03139, 2026

    TianheWu, RuibinLi, LeiZhang, andKedeMa. Diversity-preserveddistributionmatchingdistillationforfastvisual synthesis.arXiv preprint arXiv:2602.03139, 2026

    2026 arXiv

  41. [49]

    Self forcing: Bridging the train-test gap in autoregressive video diffusion.arXiv preprint arXiv:2506.08009, 2025

    Xun Huang, Zhengqi Li, Guande He, Mingyuan Zhou, and Eli Shechtman. Self forcing: Bridging the train-test gap in autoregressive video diffusion.arXiv preprint arXiv:2506.08009, 2025

    2025 arXiv

  42. [50]

    Long-horizon streaming video generation via hybrid attention with decoupled distillation.arXiv preprint arXiv:2604.10103, 2026

    Ruibin Li, Tao Yang, Fangzhou Ai, Tianhe Wu, Shilei Wen, Bingyue Peng, and Lei Zhang. Long-horizon streaming video generation via hybrid attention with decoupled distillation.arXiv preprint arXiv:2604.10103, 2026

    2026 arXiv

  43. [51]

    Rolling forcing: Autoregressive long video diffusion in real time.arXiv preprint arXiv:2509.25161, 2025

    Kunhao Liu, Wenbo Hu, Jiale Xu, Ying Shan, and Shijian Lu. Rolling forcing: Autoregressive long video diffusion in real time.arXiv preprint arXiv:2509.25161, 2025

    2025 arXiv

  44. [52]

    Omniavatar: Efficient audio-driven avatar video generation with adaptive body animation.arXiv preprint arXiv:2506.18866, 2025

    Qijun Gan, Ruizi Yang, Jianke Zhu, Shaofei Xue, and Steven Hoi. Omniavatar: Efficient audio-driven avatar video generation with adaptive body animation.arXiv preprint arXiv:2506.18866, 2025

    2025 arXiv

  45. [53]

    Hunyuanvideo-avatar: High-fidelity audio-driven human animation for multiple characters.arXiv preprint arXiv:2505.20156, 2025

    Yi Chen, Sen Liang, Zixiang Zhou, Ziyao Huang, Yifeng Ma, Junshu Tang, Qin Lin, Yuan Zhou, and Qinglin Lu. Hunyuanvideo-avatar: High-fidelity audio-driven human animation for multiple characters.arXiv preprint arXiv:2505.20156, 2025

    2025 arXiv

  46. [54]

    Evaluating prerequisite qualities for learning end-to-end dialog systems, 2016

    Jesse Dodge, Andreea Gane, Xiang Zhang, Antoine Bordes, Sumit Chopra, Alexander Miller, Arthur Szlam, and Jason Weston. Evaluating prerequisite qualities for learning end-to-end dialog systems, 2016

    2016

  47. [55]

    OpenAI. Chatgpt. https://chatgpt.com/, 2026. Accessed: 2026-07-17

    2026

  48. [56]

    Imagebind: One embedding space to bind them all

    Rohit Girdhar, Alaaeldin El-Nouby, Zhuang Liu, Mannat Singh, Kalyan Vasudev Alwala, Armand Joulin, and Ishan Misra. Imagebind: One embedding space to bind them all. InCVPR, 2023

    2023

  49. [57]

    Large- scale contrastive language-audio pretraining with feature fusion and keyword-to-caption augmentation

    Yusong Wu*, Ke Chen*, Tianyu Zhang*, Yuchen Hui*, Taylor Berg-Kirkpatrick, and Shlomo Dubnov. Large- scale contrastive language-audio pretraining with feature fusion and keyword-to-caption augmentation. InIEEE International Conference on Acoustics, Speech and Signal Processing...

    2023

  50. [58]

    Unified reward model for multimodal understanding and generation.arXiv preprint arXiv:2503.05236, 2025

    Yibin Wang, Yuhang Zang, Hao Li, Cheng Jin, and Jiaqi Wang. Unified reward model for multimodal understanding and generation.arXiv preprint arXiv:2503.05236, 2025

    2025 arXiv

  51. [59]

    Gemini 3.5 flash

    Google DeepMind. Gemini 3.5 flash. https://deepmind.google/models/gemini/flash/, 2026. Accessed: 2026-07-08

    2026

  52. [60]

    Lora: Low-rank adaptation of large language models.Iclr, 1(2):3, 2022

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Liang Wang, Weizhu Chen, et al. Lora: Low-rank adaptation of large language models.Iclr, 1(2):3, 2022. Visual Computing Lab·The Hong Kong Polytechnic University 17 / 24

    2022

  53. [61]

    Talkvid: A large-scale diversified dataset for audio-driven talking head synthesis

    Shunian Chen, Hejin Huang, Yexin Liu, Zihan Ye, Pengcheng Chen, Chenghao Zhu, Michael Guan, Rongsheng Wang, Junying Chen, Jianye Hou, et al. Talkvid: A large-scale diversified dataset for audio-driven talking head synthesis. InProceedings of the IEEE/CVF Conference on Computer...

    2026

  54. [62]

    Echomimicv2: Towards striking, simplified, and semi-body human animation.arXiv preprint arXiv:2411.10061, 2024

    Rang Meng, Xingyu Zhang, Yuming Li, and Chenguang Ma. Echomimicv2: Towards striking, simplified, and semi-body human animation.arXiv preprint arXiv:2411.10061, 2024

    2024

  55. [63]

    audio_visual_consistency

    Zhimeng Zhang, Lincheng Li, Yu Ding, and Changjie Fan. Flow-guided one-shot talking face generation with a high-resolution audio-visual dataset. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3661–3670, 2021. Visual Computing Lab·The...

    2021

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.