Pith. sign in

REVIEW 4 major objections 5 minor 45 references

AptAvatar claims that a two-step diffusion student can deliver a 60x-faster avatar video generator while preserving 720p fidelity and long-horizon identity, using an endpoint-anchored distillation target and replay of self-generated history

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-07-31 23:15 UTC pith:6GHPE2GE

load-bearing objection Solid two-step distillation paper with real techniques, but the production-readiness claim is stretched by missing wall-clock timing, a questionable mosaicked test set, and a 10-rater user study. the 4 major comments →

arxiv 2607.24013 v2 pith:6GHPE2GE submitted 2026-07-27 cs.CV

AptAvatar: Fast and Vivid Long-Form Audio-Driven Video Generation for Production-Ready Avatars

Hengyuan Zhang , Jingna Sun , Meiguang Jin , Junfeng Ma This is my paper
classification cs.CV
keywords audio-driven avatar generationdistribution matching distillationfew-step diffusionlong-form video generationendpoint-anchored distillationself-generated history replayvideo diffusion modelidentity consistency
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper tries to show that production-quality audio-driven avatar video generation need not sacrifice fidelity, resolution, or temporal window for speed. It distills a full-capacity 14B bidirectional diffusion model into a two-step generator by anchoring distribution matching to the trajectory-endpoint distribution of a frozen four-step bridge, and mitigates long-video drift by replaying cached chunks from its own earlier checkpoints as history conditions. If correct, this makes minute-scale 720p avatar generation practical at near-interactive speed without architectural compromises. The paper reports a 60x inference speedup and leads on every benchmark metric it measures.

Core claim

On its own terms, the paper's central claim is that extreme two-step distillation of a large bidirectional avatar diffusion model is stable and faithful when the student is matched not to the distant multi-step teacher distribution but to the endpoint distribution of an intermediate frozen four-step bridge generator. An Anchor Score Estimator, trained once on that fixed endpoint distribution, supplies a stationary target that complements the standard real/fake score difference; a second mechanism, Self-Generated History Replay, replaces costly online self-rollouts with a stack of detached chunks generated by past checkpoints, so the model learns to condition on its own history without backpr

What carries the argument

Endpoint-Anchored Distribution Distillation (EADD) trains an anchor score estimator on the trajectory-endpoint distribution x̂0 = G^{T→0}_φ(z,c) of a frozen four-step bridge generator; the student update uses the difference (anchor score − fake score) as a stable, attainable target instead of relying solely on the distant real-score difference. Self-Generated History Replay (SGHR) maintains a stack of detached chunks from earlier generator checkpoints and stochastically replays them as history conditions with increasing depth, approximating online self-rollout conditioning without generating multiple chunks per iteration.

Load-bearing premise

The four-step bridge generator's endpoint distribution is a faithful, unbiased target for the two-step student; if the bridge produces systematic errors, the anchor score pulls the student toward those same errors and nothing in the method provably corrects that bias.

What would settle it

Generate a video where the frozen four-step bridge exhibits clearly distorted hands; if the two-step student reproduces or amplifies those distortions rather than correcting them, the anchor-bias assumption fails. Alternatively, compare identity retention at 10 minutes against a model trained with true online rollouts: if SGHR's cached histories produce measurably worse drift, the approximation claim weakens.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • 2-NFE 720p avatar generation becomes viable for production, cutting inference cost roughly 60x relative to a 120-step baseline.
  • Endpoint-anchored distillation may transfer to other few-step video or audio-video distillation settings where teacher scores are too distant for direct supervision.
  • Self-generated history replay offers a cheaper substitute for online self-rollout training, making long-horizon consistency feasible at high resolution.
  • A full 14B-parameter model with bidirectional attention and a 3-second temporal window can be retained while reaching near-real-time inference.

Where Pith is reading between the lines

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

  • If the frozen four-step bridge carries systematic artifacts, the EADD anchor will pull the two-step student toward those artifacts; the paper's guard, the additive DMD term, is not analyzed for its ability to correct such bias.
  • The replay buffer approximates, but does not exactly match, the current generator's error distribution; differences may surface only at horizons beyond the tested minute scale.
  • The method depends on the existence of a strong four-step bridge; teams without such a checkpoint may not reproduce the reported 2-step gains.
  • A testable extension is to track identity and hand-integrity metrics at 5–10 minute horizons; if drift recurs, SGHR's coverage of degradation types is incomplete.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper presents AptAvatar, a 14B-parameter audio-driven avatar video generation framework that claims to generate vivid 720p long-form videos with only 2 NFE (function evaluations), a 60x speedup over the multi-step InfiniteTalk baseline. Two technical contributions are proposed: Endpoint-Anchored Distribution Distillation (EADD), which trains an anchor score estimator on the endpoint distribution of a frozen 4-step bridge generator to stabilize two-step distillation, and Self-Generated History Replay (SGHR), which uses a memory bank of cached self-generated chunks to mitigate exposure bias during long-form chunk-wise generation. The authors report state-of-the-art quantitative results on short-video and minute-level long-video benchmarks, plus a user study.

Significance. If the claims are validated, this would be a substantial practical contribution: two-step generation at 720p with a full-capacity bidirectional model, a 3-second temporal window, and long-horizon consistency would remove a major deployment bottleneck for production avatar systems. The paper's strengths include explicit loss equations, ablations of both proposed components, evaluation on standard external benchmarks (FID, VBench, Sync-C), and a promise of code release. However, several evaluation gaps and a theoretical risk in the EADD design need to be addressed before the headline claims are established.

major comments (4)
  1. [§4.1 Datasets] The first test set is described as '100 videos of 15 seconds each, with faces mosaicked due to copyright issues.' If the mosaicking applies to the evaluation videos, then face-dependent metrics — especially Sync-C/Sync-D (lip-audio synchronization), identity preservation, and arguably ASE (aesthetics) — cannot be meaningfully computed, since the face region is destroyed. This directly undermines the central claim of 'preserving visual fidelity and long-horizon identity.' Please clarify whether the mosaicking applies only to the source/reference videos, and if not, re-run the face-related metrics on unmosaicked evaluation data or explain why the metrics remain valid.
  2. [§1 and §4.2 (60x speedup)] The 60x speedup is presented as a ratio of NFE counts (120 for InfiniteTalk vs. 2 for AptAvatar), not a measured wall-clock latency. Inference time also includes VAE encoding/decoding, text/audio conditioning, and the memory-bank bookkeeping of SGHR. For a production-readiness claim, please report actual end-to-end latency on the same GPU(s), with batch size and hardware specified, and compare against the 4-step methods (LiveAvatar, SoulX-FlashTalk) as well. The NFE ratio alone is not evidence of a comparable speedup in practice.
  3. [§3.2, Eqs. (4)-(6)] EADD's anchor score is trained to predict the endpoint distribution of the frozen 4-step bridge generator G_phi. If G_phi has systematic artifacts (e.g., distorted hands, blurry motion) — a failure mode the paper itself attributes to few-step generation — then the anchor will pull the 2-step student toward those artifacts. The DMD term in Eq. (6) is supposed to correct this via the real score, but the paper's own motivation is that the real score is unreliable for out-of-distribution student samples, i.e., precisely in the regime where the anchor is most needed. The ablation in Table 2 reports aggregate metrics and the user study compares against other methods, not against the 4-step bridge. Please provide a direct comparison between the 4-step bridge and the 2-step student on artifact-specific measures (e.g., hand-keypoint integrity, identity embedding distance) or an analysis showing t
  4. [§4.2 User study and Tables 1-2] The user study uses only 10 raters and reports a single aggregated win rate (62%) without confidence intervals or inter-rater agreement. The quantitative tables also lack error bars or significance tests; several reported differences between methods are very small (e.g., Sync-C 7.865 vs. 7.743, Motion-S 99.561 vs. 99.550 in the short-video benchmark). Given the 'state-of-the-art' claim, please report variance across the 100/20 test videos, number of raters per video, and statistical significance (or at least confidence intervals) for the key metrics. As reported, the differences may not be distinguishable from noise.
minor comments (5)
  1. [Table 1] FID is reported as '–' for all long-video rows without explanation. Please clarify whether FID is not computed on the 1-minute benchmark and why.
  2. [§3.3, Eq. (7)] The notation '(K+1)-th chunk' is confusing when K is sampled from the memory bank and is also used as a depth counter. Please define the replay depth more explicitly and explain how the memory context m' is updated when a cached chunk becomes the new history.
  3. [Figure 2] The captions for the EADD branches ("EADD: Generator Update" and "EADD: Critic Update") are terse; it would help to state which score estimators are trainable/frozen in each branch.
  4. [§4.2 User study] The acronym 'GSB' is not defined; please spell it out on first use.
  5. [References] Magic1-For-1 is cited as 'Withdrawn preprint'; if it is withdrawn, consider citing a stable version or noting why it is included.

Circularity Check

0 steps flagged

No significant circularity: the central claims are evaluated against external benchmarks and do not reduce to the paper's own definitions.

full rationale

I walked the derivation chain for EADD and SGHR and found no step where a 'prediction' is equivalent by construction to an input or fitted parameter. The Anchor Score Estimator in Eq. (4) is trained on the endpoint distribution of the frozen four-step bridge, and the EADD gradient in Eq. (5) uses that anchor together with the fake-score estimator. This is an internal training target, but the paper's headline claims (2-NFE fidelity, long-horizon identity, 60x speedup) are measured with external metrics: FID, VBench Subject-C/BG-C/Motion-S/Temporal-F, ASE, Sync-C/Sync-D, and a GSB user study. Those evaluations are computed on real video data and compare against other published methods, so the central result could fail independently of the training objective. The ablation in Table 2 (Ours w/o ASE) shows that removing the anchor degrades these external metrics, which is meaningful evidence that the anchor is not a vacuous rename of the output. SGHR reuses cached chunks from earlier generator checkpoints as history conditions; this is a training-data augmentation strategy, not a fitted parameter renamed as a prediction, and it is compared against online-rollout baselines. No self-citations carrying the argument were found, no uniqueness theorem imported from the authors is invoked, and no ansatz is smuggled in via a citation to prior work that itself merely adopts the ansatz. The skeptic's concern about the anchor inheriting bridge artifacts is a correctness/robustness caveat, not a circularity: the reported metrics could in principle reveal such artifacts, and the paper's methods are externally falsifiable. Therefore the appropriate circularity score is 0.

Axiom & Free-Parameter Ledger

4 free parameters · 3 axioms · 2 invented entities

The method depends on several hyperparameters that are not reported (lambda_eadd, p_rep, K_max), on the validity of DMD from prior work, and on the untested assumption that a 4-step bridge is an unbiased anchor. The two new components (anchor score, memory bank) are internal engineering mechanisms with evidence only from the paper's own ablations.

free parameters (4)
  • lambda_eadd = not reported
    Weight of the EADD loss in Eq. (6); the paper never states its value, though the final result depends on it.
  • p_rep = not reported
    Probability of sampling history context from the replay stack in Eq. (7); controls how much the model sees degraded histories.
  • K_max = not reported
    Maximum replay depth before reset; controls how long self-generated history chains can be.
  • bridge steps = 4
    The intermediate generator is fixed at four denoising steps; this is a hand-chosen schedule that defines the anchor distribution.
axioms (3)
  • standard math DMD gradient (Eq. 2) with real and fake scores correctly reduces distribution divergence for a few-step student.
    Relies on established DMD theory from the cited literature; the paper uses it without proof.
  • domain assumption The frozen four-step bridge's endpoint distribution is a stable, attainable target for two-step distillation (Section 3.2).
    Key assumption of EADD: matching a 4-step model's output is easier and does not introduce bias. This is not proven; the ablation is the only evidence.
  • domain assumption Replaying cached chunks from old checkpoints is a sufficient substitute for online self-rollouts (Section 3.3).
    SGHR assumes the distribution of past-generated degradations in the memory bank approximates inference-time self-conditioning. No analysis, only empirical comparison.
invented entities (2)
  • Anchor Score Estimator (sa) no independent evidence
    purpose: Learned score network trained on the four-step bridge's endpoint distribution; provides a stable target for the two-step student in Eq. (5).
    Its benefit is demonstrated only by the paper's ablation (Ours vs. Ours w/o ASE). No external falsifiable prediction.
  • Memory Bank (SGHR) no independent evidence
    purpose: Stack-structured replay buffer of detached chunks from earlier generator checkpoints; used as history conditions in Eq. (7)-(8).
    No externally testable prediction; it is validated only via the long-video metrics in Table 2.

pith-pipeline@v1.3.0-alltime-deepseek · 11604 in / 16340 out tokens · 135072 ms · 2026-07-31T23:15:37.958422+00:00 · methodology

0 comments
read the original abstract

Production-ready audio-driven avatar generation requires efficient inference without sacrificing fidelity or motion expressiveness. However, existing acceleration methods often compromise quality through restrictive architectural choices, such as causal attention and short temporal horizons, or by reducing model capacity and resolution. Without such compromises, we propose AptAvatar, a 14B-parameter long-form audio-driven avatar generation framework that delivers fast and expressive inference. For efficiency in production-level applications, AptAvatar addresses the extreme two-step generation challenge. To bridge the gap between the multi-step teacher model and the two-step student model, we introduce Endpoint-Anchored Distribution Distillation. It augments vanilla distribution matching with a dedicated Anchor Score Estimator trained on the trajectory-endpoint distribution defined from a frozen pretrained 4-step bridge generator. This provides an attainable endpoint-level anchor for the evolving two-step student. To improve long-horizon consistency, we further introduce Self-Generated History Replay, which reuses cached outputs from earlier generator checkpoints as history conditions during chunk-wise training. This approximates inference-time conditioning on self-generated histories without costly online rollouts, mitigating quality degradation from accumulated history errors. Extensive experiments demonstrate that AptAvatar generates vivid 720p long-form avatar videos with only 2 NFEs, achieving a 60x speedup while preserving visual fidelity and long-horizon identity. Code is available at https://github.com/TaoLiveAIGC/AptAvatar

Figures

Figures reproduced from arXiv: 2607.24013 by Hengyuan Zhang, Jingna Sun, Junfeng Ma, Meiguang Jin.

Figure 1
Figure 1. Figure 1: Teaser of AptAvatar. Given a reference identity, speech audio, and action prompts, AptAvatar generates 720p avatar videos with natural audio-driven motion and support for fine-grained text control over specific actions and interactions. Using only 2 NFEs, AptAvatar preserves vivid motion, identity consistency, and stable long-form synthesis, achieving a 60× inference speedup for production-ready avatar app… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of AptAvatar. (a) The overall chunk-wise framework encodes audio, text, reference image, and video history conditions, then generates the next chunk with a two-step DiT generator. EADD supervises two-step distillation, and SGHR updates the history conditions used during training. (b) EADD introduces an anchor score estimator learned from the endpoint distribution of a frozen four-step bridge gener… view at source ↗
Figure 3
Figure 3. Figure 3: Fine-grained text-guided motion generation. AptAvatar faithfully follows action-specific textual prompts and produces [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Visual comparison on short-video and minute-level long-video test sets. AptAvatar preserves hand integrity, identity [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Human preference win rate under the GSB protocol. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗

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

45 extracted references · 2 linked inside Pith

  1. [1]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Audcast: Audio-driven human video generation by cascaded diffusion transformers , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  2. [2]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Transition Matching Distillation for Fast Video Generation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  3. [3]

    2026 , eprint=

    One-Forcing: Towards Stable One-Step Autoregressive Video Generation , author=. 2026 , eprint=

    2026

  4. [4]

    Proceedings of the 43rd International Conference on Machine Learning , volume=

    Optimizing Few-Step Generation with Adaptive Matching Distillation , author=. Proceedings of the 43rd International Conference on Machine Learning , volume=

  5. [5]

    2025 , eprint=

    InfiniteTalk: Audio-Driven Video Generation for Sparse-Frame Video Dubbing , author=. 2025 , eprint=

    2025

  6. [6]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    From slow bidirectional to fast autoregressive video diffusion models , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  7. [7]

    Proceedings of the 43rd International Conference on Machine Learning , volume=

    Causal Forcing: Autoregressive Diffusion Distillation Done Right for High-Quality Real-Time Interactive Video Generation , author=. Proceedings of the 43rd International Conference on Machine Learning , volume=

  8. [8]

    Advances in Neural Information Processing Systems , volume=

    Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion , author=. Advances in Neural Information Processing Systems , volume=

  9. [9]

    The Fourteenth International Conference on Learning Representations , year=

    Self-Forcing++: Towards Minute-Scale High-Quality Video Generation , author=. The Fourteenth International Conference on Learning Representations , year=

  10. [10]

    2026 , eprint=

    High-Fidelity Two-Step Image Generation via Teacher-Aligned End-to-End Distillation , author=. 2026 , eprint=

    2026

  11. [11]

    Wu, Zhuguanyu and Gong, Ruihao and Yong, Yang and Huang, Yushi and Fan, Xiangyu and Yang, Lei and Lin, Dahua and Liu, Xianglong , booktitle=

  12. [12]

    Advances in Neural Information Processing Systems , volume=

    Improved Distribution Matching Distillation for Fast Image Synthesis , author=. Advances in Neural Information Processing Systems , volume=

  13. [13]

    2026 , eprint=

    TurboTalk: Progressive Distillation for One-Step Audio-Driven Talking Avatar Generation , author=. 2026 , eprint=

    2026

  14. [14]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Phased one-step adversarial equilibrium for video diffusion models , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  15. [15]

    Li, Haobo and Zeng, Yanhong and Lu, Yunhong and Zhu, Jiapeng and Ouyang, Hao and Wang, Qiuyu and Cheng, Ka Leong and Shen, Yujun and Zhang, Zhipeng , booktitle=

  16. [16]

    Transactions on Machine Learning Research , year=

    ConsistI2V: Enhancing Visual Consistency for Image-to-Video Generation , author=. Transactions on Machine Learning Research , year=

  17. [17]

    2024 , eprint=

    Tuning-Free Noise Rectification for High Fidelity Image-to-Video Generation , author=. 2024 , eprint=

    2024

  18. [18]

    2025 , eprint=

    Magic 1-For-1: Generating One Minute Video Clips within One Minute , author=. 2025 , eprint=

    2025

  19. [19]

    European conference on computer vision , pages=

    Freeinit: Bridging initialization gap in video diffusion models , author=. European conference on computer vision , pages=. 2024 , organization=

    2024

  20. [20]

    2026 , eprint=

    SoulX-FlashTalk: Real-Time Infinite Streaming of Audio-Driven Avatars via Self-Correcting Bidirectional Distillation , author=. 2026 , eprint=

    2026

  21. [21]

    Machine learning , volume=

    Self-improving reactive agents based on reinforcement learning, planning and teaching , author=. Machine learning , volume=. 1992 , publisher=

    1992

  22. [22]

    nature , volume=

    Human-level control through deep reinforcement learning , author=. nature , volume=. 2015 , publisher=

    2015

  23. [23]

    Proceedings of the 28th ACM international conference on multimedia , pages=

    A lip sync expert is all you need for speech to lip generation in the wild , author=. Proceedings of the 28th ACM international conference on multimedia , pages=

  24. [24]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Sadtalker: Learning realistic 3d motion coefficients for stylized audio-driven single image talking face animation , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  25. [25]

    European Conference on Computer Vision , pages=

    Emo: Emote portrait alive generating expressive portrait videos with audio2video diffusion model under weak conditions , author=. European Conference on Computer Vision , pages=. 2024 , organization=

    2024

  26. [26]

    The Thirteenth International Conference on Learning Representations , year=

    Cyberhost: A one-stage diffusion framework for audio-driven talking body generation , author=. The Thirteenth International Conference on Learning Representations , year=

  27. [27]

    Advances in Neural Information Processing Systems , volume=

    Let Them Talk: Audio-Driven Multi-Person Conversational Video Generation , author=. Advances in Neural Information Processing Systems , volume=

  28. [28]

    2508.18621 , archivePrefix=

    Gao, Xin and Hu, Li and Hu, Siqi and Huang, Mingyang and Ji, Chaonan and Meng, Dechao and Qi, Jinwei and Qiao, Penchong and Shen, Zhen and Song, Yafei and others , year=. 2508.18621 , archivePrefix=

    Pith/arXiv arXiv

  29. [29]

    2505.20156 , archivePrefix=

    Chen, Yi and Liang, Sen and Zhou, Zixiang and Huang, Ziyao and Ma, Yifeng and Tang, Junshu and Lin, Qin and Zhou, Yuan and Lu, Qinglin , year=. 2505.20156 , archivePrefix=

    Pith/arXiv arXiv

  30. [30]

    Seminal Graphics Papers: Pushing the Boundaries, Volume 2 , pages=

    A morphable model for the synthesis of 3D faces , author=. Seminal Graphics Papers: Pushing the Boundaries, Volume 2 , pages=

  31. [31]

    , author=

    Learning a model of facial shape and expression from 4D scans. , author=. ACM Trans. Graph. , volume=

  32. [32]

    2026 , eprint=

    LongCat-Video-Avatar 1.5 Technical Report , author=. 2026 , eprint=

    2026

  33. [33]

    International Conference on Learning Representations , volume=

    One step diffusion via shortcut models , author=. International Conference on Learning Representations , volume=

  34. [34]

    International Conference on Learning Representations , volume=

    Simplifying, stabilizing and scaling continuous-time consistency models , author=. International Conference on Learning Representations , volume=

  35. [35]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    One-step diffusion with distribution matching distillation , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  36. [36]

    Liu, Dongyang and Gao, Peng and Liu, David and Du, Ruoyi and Li, Zhen and Wu, Qilong and Jin, Xin and Cao, Sihan and Zhang, Shifeng and Li, Hongsheng and Hoi, Steven C. H. , booktitle=. Decoupled

  37. [37]

    Proceedings of the 42nd International Conference on Machine Learning , pages=

    Diffusion Adversarial Post-Training for One-Step Video Generation , author=. Proceedings of the 42nd International Conference on Machine Learning , pages=

  38. [38]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Learning few-step diffusion models by trajectory distribution matching , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  39. [39]

    European Conference on Computer Vision , year=

    Distribution Matching Distillation Meets Reinforcement Learning , author=. European Conference on Computer Vision , year=

  40. [40]

    Advances in Neural Information Processing Systems , volume=

    Frame Context Packing and Drift Prevention in Next-Frame-Prediction Video Diffusion Models , author=. Advances in Neural Information Processing Systems , volume=

  41. [41]

    The Fourteenth International Conference on Learning Representations , year=

    LongLive: Real-Time Interactive Long Video Generation , author=. The Fourteenth International Conference on Learning Representations , year=

  42. [42]

    2026 , eprint=

    Live Avatar: Streaming Real-Time Audio-Driven Avatar Generation with Infinite Length , author=. 2026 , eprint=

    2026

  43. [43]

    Q-Align: Teaching

    Wu, Haoning and Zhang, Zicheng and Zhang, Weixia and Chen, Chaofeng and Liao, Liang and Li, Chunyi and Gao, Yixuan and Wang, Annan and Zhang, Erli and Sun, Wenxiu and Yan, Qiong and Min, Xiongkuo and Zhai, Guangtao and Lin, Weisi , booktitle=. Q-Align: Teaching

  44. [44]

    Asian conference on computer vision , pages=

    Out of time: automated lip sync in the wild , author=. Asian conference on computer vision , pages=. 2016 , organization=

    2016

  45. [45]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Vbench: Comprehensive benchmark suite for video generative models , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=