Pith. sign in

REVIEW 5 major objections 6 minor 43 references

LiveAnimate: Stable Long-Form Streaming Human Animation in Real-Time

T0 review · 5 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 LiveAnimate turns a 14B video diffusion transformer into a real-time streaming animator that keeps identity and quality nearly flat over three minutes, at about 20 FPS on two H100 GPUs.

desk verdict Strong engineering and a genuinely new operating point, but 'real-time' and 'arbitrarily long' are both slightly ahead of the evidence. read the letter →

arxiv 2608.11745 v2 pith:R6Y4DHIW submitted 2026-08-12 cs.CV

Yuxuan Zhang , Haozhong Xiong , Yubo Huang , Jiayi Song , Jinpeng Yu , Haofan Wang , Jiaming Liu , Ruihua Huang
show 1 more author
Liwei Wang
This is my paper · ORCID
classification cs.CV
keywords pose-drivenhumananimationstreamingvideogenerationdiffusiontransformerreal-timeinferenceKVcacheattentionsinkself-forcingdistillationlong-formstability
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 claims that pose-driven human animation can be made simultaneously real-time, streaming, and stable over long rollouts by rebuilding a billion-scale video diffusion transformer into a block-causal generator and giving it a bounded memory that recalls pose-relevant history. The authors' system, LiveAnimate, converts a pretrained bidirectional 14B-parameter DiT into an autoregressive block generator and distills it to three denoising steps, then adds a fixed-size cache called Pose-Retrieval Sink Attention that anchors the first generated block and retrieves a historical pose-matched block when poses recur. On a three-minute benchmark it reports nearly flat perceptual quality (IQA 4.047 to 4.026) and identity (DINO-S 0.833 to 0.818) while generating at 19.63 FPS on two H100 GPUs, with per-block latency and memory independent of stream length. If right, this moves full-body animation from an offline, minutes-to-hours process to an interactive one, making live streaming, telepresence, and virtual avatars practical with diffusion-quality output.

What carries the argument

The load-bearing mechanism is Pose-Retrieval Sink Attention (PR-Sink), a bounded cache that gives every attention layer four fixed regions: the global Ref Sink holding reference-image latents computed once at the clean timestep, a permanent Static Sink holding the first generated block, a Dynamic Sink holding a historical block retrieved by pose similarity, and a three-slot Rolling Window of recent clean blocks plus the current block. The Dynamic Sink is selected from a memory bank of five entries keyed by a 558-dimensional whole-body pose fingerprint (133 keypoints across three frames, with confidence scores and $\ell^2$ normalization); bank updates are designed to maximize pose coverage but stop after the first 20 blocks. Retrieved keys and values are cached before rotary position embeddings and re-rotated at attention time, which lets the same bank entry be reused at any later position. Each generated block is written into the window via a Clean KV Update, and because every region has fixed capacity, attention cost and memory stay constant as the stream grows.

What would settle it

Run LiveAnimate on a ten-minute driving stream constructed so that no pose after the first 20 blocks has high cosine similarity to any of the five bank entries, for example a choreography that cycles through a large pose vocabulary without repeating itself, and measure DINO-S on the final segment; if identity drops toward or below the reported w/o-dynamic-sink final value of 0.805 while frame-level IQA stays high, the bounded PR-Sink has stopped supplying pose-relevant context and the stable-long-form claim fails for novel-pose streams.

Watch

Extended reading notes

Core claim

The paper's central claim is that the three coupled obstacles to interactive human animation—reducing a 14B diffusion transformer to an interactive latency budget, converting a bidirectional model into a causal generator, and preventing identity drift over open-ended rollouts—can be solved together. LiveAnimate takes a pretrained bidirectional 14B-parameter video DiT and re-trains it in two stages: Reference-Anchored Teacher-Forcing Adaptation makes it block-causal while keeping the reference image visible as a permanent Ref Sink, and Block-wise Self-Forcing Distillation reduces sampling to three denoising steps by replaying one block at a time through a distribution-matching loss. For long streams it uses Pose-Retrieval Sink Attention (PR-Sink), a bounded KV cache that keeps the first generated block as a Static Sink, retrieves a pose-matched historical block from a five-entry bank into a Dynamic Sink, and slides a three-block Rolling Window, with all cached keys re-rotated so they can be reused at any position. The authors report that this yields 19.63 FPS on two H100 GPUs with per-block latency of 0.611 seconds, constant regardless of stream length, and that on a three-minute benchmark IQA falls only from 4.047 to 4.026 and DINO-S from 0.833 to 0.818 while baselines degrade or require hours of offline computation.

Load-bearing premise

The long-form stability claim assumes that poses appearing late in the stream resemble poses from the first 20 blocks closely enough that the fixed five-entry pose bank can retrieve a matching historical block; a stream of continuously novel poses would leave the Dynamic Sink stale and remove its identity-preserving benefit.

Editorial extensions

If this is right

  • Memory and per-block latency stay constant regardless of stream duration, since every cache region has fixed capacity; the recorded 0.611 seconds per 12-frame block does not grow over minutes.
  • Three-step sampling with block-wise self-forcing distillation is enough to keep quality close to the teacher-forcing baseline; four steps improve frame-level scores slightly but add an extra denoising pass, while two steps weaken long-horizon identity.
  • On the three-minute benchmark, LiveAnimate is the only evaluated method that keeps perceptual quality and identity nearly flat; competing systems either accumulate visible degradation or need hours of offline compute for the same rollout.
  • The Static Sink and Dynamic Sink play complementary roles: removing the Static Sink collapses final-segment DINO-S to 0.693, while removing the Dynamic Sink drops it to 0.805, confirming that a permanent identity anchor and pose-relevant retrieval are both needed.
  • Two-GPU Ulysses sequence parallelism is the chosen operating point, giving a 1.58x speedup at 79.1% efficiency; four GPUs saturate at 1.78x and are not worth the added communication.

Reading between the lines

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

  • The fixed 5-entry bank with updates frozen after block 20 implies that the method's promise is strongest for pose-repetitive or slowly-changing streams; a live session that keeps introducing genuinely novel articulations after block 20 would likely degrade toward the w/o-dynamic-sink ablation. The paper's benchmark, which repeats each pose forward-reverse-forward, does not exercise that regime.
  • A testable extension is to make bank updates continue throughout the stream or grow the bank adaptively, which could extend stability to non-repetitive choreography; the current design trades freshness for bounded memory.
  • Because cached keys are re-rotated at attention time, the same retrieved block can be reused at arbitrary later timestamps, so the PR-Sink mechanism should generalize to other retrieval-conditioned streaming video tasks such as scene revisits or camera loopbacks.
  • Since denoising accounts for 75% of per-block time and the Clean KV Update for 24%, a further speedup would likely come from predicting clean KV states directly rather than running an extra forward pass at the clean timestep.
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

5 major / 6 minor

Summary. This paper presents LiveAnimate, a streaming pose-driven human animation system built on a 14B-parameter video Diffusion Transformer. The authors propose a two-stage training pipeline: Reference-Anchored Teacher-Forcing Adaptation converts a pretrained bidirectional DiT into a block-causal generator, and Block-wise Self-Forcing Distillation reduces sampling to three steps via a one-block-at-a-time replay scheme. For long-form stability, they introduce Pose-Retrieval Sink Attention (PR-Sink), a bounded KV cache combining a static sink, a pose-retrieved dynamic sink, and a rolling window. On a three-minute benchmark, they report nearly flat trajectories for aesthetic score, no-reference IQA, and DINO similarity, with 19.63 FPS on two H100 GPUs, and claim the first real-time streaming system at billion scale for stable long-form animation.

Significance. If the claims hold, LiveAnimate would be a notable engineering contribution: it is the first system to operate a 14B-parameter pose-driven animation model at interactive rates over extended streams. The PR-Sink mechanism is an elegant bounded-cache design for reusing pose-relevant historical context, and the block-wise self-forcing distillation recipe is practical for single-node distillation of an autoregressive video model. The flat metric trajectories over three minutes are encouraging. However, the central claims of 'real-time' and 'stable long-form generation' are not yet fully supported by the present evidence: the reported throughput is below the benchmark's 25 FPS, the latency excludes VAE operations, and the long-form evaluation is constructed so that poses recur within the fixed bank-update horizon. The system, as described, is a plausible foundation, but the evaluation must be tightened before the claims can be accepted.

major comments (5)
  1. [Sec. 4.1 (Latency and throughput measurement)] The headline throughput of 19.63 FPS measures only the DiT generation loop, excluding VAE encoding/decoding and condition processing. Since the benchmark videos are 25 FPS, the reported throughput is below the input frame rate, so the 'real-time streaming' claim in the abstract and Sec. 1 is not supported by the measured numbers. Please report end-to-end latency including VAE operations, or explicitly qualify the claim as 'DiT-loop real-time' and quantify the pipeline overhead.
  2. [Sec. 3.4 (Bank update restriction)] The paper restricts bank updates to the first 20 blocks, which at 12 RGB frames per block and 25 FPS corresponds to approximately 9.6 seconds of a three-minute stream. This means the Dynamic Sink can only retrieve historical context from poses observed in the first 9.6 seconds. For an open-ended stream that introduces novel poses after block 20, the method degrades toward the w/o-dynamic-sink ablation, whose final-segment DINO-S drops to 0.805 (Fig. 7). The controlled benchmark uses forward-reverse-forward sequences, so every later pose already appears within the first 20 blocks; the in-the-wild half is aggregated without per-sequence breakdown. The claim of 'stable long-form generation' for 'arbitrarily long rollouts' is therefore not demonstrated for streams with novel poses after the update horizon.
  3. [Sec. 4.2 and Fig. 5] The reported temporal differences are very small (e.g., IQA from 4.047 to 4.026, DINO-S from 0.833 to 0.818), and the paper provides no error bars, confidence intervals, or number of independent runs. It is unclear whether these differences are statistically significant or within run-to-run variance. Please provide variance estimates or per-sequence breakdowns for both the controlled and in-the-wild halves of the benchmark.
  4. [Sec. 4.1 and Fig. 5] The comparison with baselines is not on a consistent basis. LiveAnimate's runtime of approximately 4 minutes for three minutes of video is computed from the DiT generation loop only, while the baselines' reported 2-5 hours appear to be end-to-end wall-clock time. The same measurement protocol should be applied to all methods, with a clear statement of which components are included in each timing.
  5. [Sec. 3.4 and Sec. 4.4] The bank capacity M=5 and the restriction of updates to the first 20 blocks are design choices that are neither justified nor ablated. Since these parameters directly determine whether the method can handle open-ended novel poses, the paper should include a sensitivity study for M and the update horizon, or explicitly discuss the intended operating regime and its limitations.
minor comments (6)
  1. [Abstract and Sec. 1] The phrase 'arbitrarily long rollouts' overstates the evidence; the longest experiment is three minutes. Please qualify this claim to match the evaluated duration.
  2. [Sec. 4.1] The text uses 'V AE' with a space; it should be 'VAE'.
  3. [Sec. 4.1, reference [8]] The citation for the 40k talking-video dataset points to a paper on audio-visual speech separation (Ephrat et al.), which appears mismatched. Please verify the correct dataset reference.
  4. [Figure 7] The 'w/o RoPE follow' variant is not defined in the main text. Please explain what this ablation removes and how it differs from the full 'Position-consistent KV reuse' described in Sec. 3.4.
  5. [Sec. 4.1] Please specify how FID is computed on temporal segments, including the number of frames used and the reference distribution, since FID is typically defined on image sets.
  6. [Table 3] The component names 'Sink read' and 'Sink write' differ from the text's 'PR-Sink retrieval and bank maintenance'; please unify the terminology.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: LiveAnimate's claims are empirical system results, and PR-Sink's bounded-memory property is a stated design consequence rather than a fitted prediction.

full rationale

LiveAnimate is an empirical systems paper. The headline numbers (19.63 FPS, IQA 4.047 to 4.026, DINO-S 0.833 to 0.818) are measured against external baselines, not derived from a fitted parameter that is then renamed as a prediction. The bounded-cache property of PR-Sink ('cache storage and attention cost do not grow with stream duration') follows directly from the fixed-capacity definitions in Sec. 3.4, and the paper explicitly presents it as a design consequence rather than as a fitted result. There are no load-bearing self-citations: the external references such as Self Forcing [16], DMD [37], attention sinks [34], and Ulysses [9] are independent prior work, and no uniqueness theorem from the authors is invoked to force a design choice. Hyperparameters such as M=5, the 20-block update cutoff, and three sampling steps are tuning choices; they are ablated but not relabeled as predictions. The X-Dance benchmark's forward-reverse-forward construction and the 20-block bank freeze could under-test streams with novel late poses, but this is an evaluation-scope limitation, not a circular reduction: the reported quality and identity trajectories are not equal to the benchmark's pose repetition by construction. No step in the paper reduces to its own input under the stated circularity criteria.

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

The central claims rest on a learned 14B base model, on a pose-fingerprint retrieval metric, on the possibility of distilling to 3 steps, on an overlap assumption for VAE costs, and on several hand-picked cache hyperparameters. None of these has independent validation beyond the reported benchmark.

free parameters (5)
  • Denoising steps = 3
    Selected from an ablation on the three-minute benchmark; 2 steps weaken long-horizon identity and 4 steps improve quality but slow inference. This choice directly produces the 19.63 FPS number.
  • Pose memory bank capacity M = 5
    Fixed capacity with no sensitivity study; determines how much pose-specific appearance context can be recalled at any later time.
  • Bank update freeze horizon = first 20 blocks
    Bank entries stop updating after 20 blocks; there is no analysis for streams where novel poses appear later.
  • Rolling window slots = 3 blocks
    Design choice for the bounded KV cache; no window-size ablation is reported.
  • LoRA rank = 128
    Fine-tuning rank for both training stages; affects trainability and capacity, with no sweep reported.
assumptions (5)
  • domain assumption Cached pre-RoPE keys can be re-rotated to a different temporal position at retrieval without changing their content semantics.
    Invoked in Sec. 3.4 'Position-consistent KV reuse'. If false, the Dynamic Sink injects stale positional information and retrieval fails.
  • domain assumption Cosine similarity of the 558-D ViTPose fingerprint is a sufficient relevance signal for pose-recurrence retrieval.
    Sec. 3.4 defines retrieval as arg max dot product over fingerprints; the paper does not validate this metric against alternatives.
  • domain assumption A pretrained bidirectional 14B DiT can be converted to a causal 3-step generator by LoRA teacher forcing plus blockwise DMD without unacceptable quality loss.
    Secs. 3.2 and 3.3. The entire system depends on this transfer, and only ablation evidence supports it.
  • domain assumption VAE encoding and decoding plus condition processing can be overlapped on separate devices, so excluding them does not change the real-time verdict.
    Sec. 4.1 latency protocol. No end-to-end wall-clock measurement is reported.
  • domain assumption A clean-timestep reference latent stored as a permanent Ref Sink remains a valid identity anchor for the whole stream.
    Sec. 3.2, Eq. (1). If the reference anchor degrades, identity stability fails even with perfect sinks.

how reviews work

0 comments
Cite this review

Pith. "Pith review of LiveAnimate: Stable Long-Form Streaming Human Animation in Real-Time." pith.science (2026). https://pith.science/paper/R6Y4DHIW

@misc{pith2026260811745,
  author       = {Pith},
  title        = {Pith review of: LiveAnimate: Stable Long-Form Streaming Human Animation in Real-Time},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/R6Y4DHIW}},
  note         = {Machine review of arXiv:2608.11745}
}
read the original abstract

Pose-driven human animation synthesizes a video of a target person from a single reference image and a driving pose stream. Real-time generation is essential for interactive applications such as live streaming, telepresence, and virtual avatars, yet diffusion-based systems require minutes to hours per clip, precluding responsive interaction. We present LiveAnimate, to our knowledge the first animation system to combine real-time streaming with stable long-form generation at billion scale, built on a 14B-parameter video Diffusion Transformer (DiT). A two-stage training pipeline first adapts a pretrained bidirectional DiT into a block-causal autoregressive generator through Reference-Anchored Teacher-Forcing Adaptation, and then reduces the sampling budget to three steps through Block-wise Self-Forcing Distillation. To preserve appearance over extended streams, we introduce Pose-Retrieval Sink Attention (PR-Sink), a bounded KV-cache mechanism combining a Static Sink that permanently anchors the first generated block, a Dynamic Sink that holds a pose-retrieved historical block, and a three-slot Rolling Window. When a pose recurs, PR-Sink restores the relevant appearance context without retaining the entire sequence, so memory and per-block latency remain constant regardless of stream duration. Together with Ulysses sequence parallelism and operator fusion, these designs enable 19.63\,FPS streaming inference on two NVIDIA H100 GPUs. On a three-minute benchmark, LiveAnimate maintains nearly constant perceptual quality and identity from the first 30 seconds to the final minute, while prior systems degrade substantially or require hours of offline computation for the same rollout. These results establish a new operating point in quality, latency, and duration for interactive full-body animation.

Figures

Figures reproduced from arXiv: 2608.11745 by the authors.

Figure 1
Figure 1. LiveAnimate enables real-time, stable long-form streaming human animation. Given a reference image and a stream of body-pose and facial controls, LiveAnimate generates identity-consistent animation causally, one temporal block at a time. Pose-Retrieval Sink Attention recalls pose-relevant historical context when similar poses recur, preserving appearance over extended streams. With three￾step sampling, LiveAnimate r… view at source ↗
Figure 2
Figure 2. Overview of LiveAnimate. Given a reference image and streaming pose signals, our system generates video blocks autoregres￾sively. Each block undergoes 3-step denoising followed by a clean KV update. PR-Sink augments a three-block rolling window with the first generated block and a pose-matched historical block selected from a compact memory bank. Ulysses sequence parallelism distributes attention computation across … view at source ↗
Figure 3
Figure 3. Qualitative comparison on a full-body sequence. Frames are sampled every 20 seconds from a three-minute, 25-FPS rollout. The pose signal is shown above the generated frames, and red annotations highlight representative long-horizon artifacts in competing methods. The rightmost column reports end-to-end generation time: the baselines require approximately 2–5 hours, whereas LiveAnimate completes the sequence in appro… view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Qualitative comparison on an upper-body sequence. Frames are sampled every 20 seconds from a three-minute rollout under the same reference image and driving-pose sequence. LiveAnimate maintains the subject’s identity, clothing, and dark background more consistently ove…
Figure 5
Figure 5. Figure 5: Quality and identity over a three-minute rollout. Metrics are computed on a 0–10 s prefix, the 0–30 s initial window, and three subsequent temporal segments. The full LiveAnimate model is highlighted in red. Flat trajectories indicate resistance to accumulated degradat…
Figure 6
Figure 6. Figure 6: Qualitative ablation over a three-minute rollout. Frames are sampled every 40 seconds from 20 to 180 seconds un￾der the same reference image and driving-pose sequence. timestep (t=0) and stores the resulting KV state for sub￾sequent blocks. PR-Sink retrieval and bank m…
Figure 7
Figure 7. Figure 7: Quantitative ablation over the three-minute rollout. Top: component and cache variants, including separate removal of the static and dynamic sinks. The w/o-DMD variant replaces DMD-based self-forcing distillation with conventional teacher forcing. Bottom: the sampling-…
Figure 8
Figure 8. Figure 8: Additional qualitative results on four three-minute sequences. For each example, the left column shows the reference image, while the rows on the right show the driving pose signals and corresponding outputs sampled every 20 seconds from 20 to 180 seconds. The examples…
Figure 9
Figure 9. Figure 9: Additional qualitative results on three three-minute sequences. We show the reference image, driving pose signal, and generated outputs at 20-second intervals. The sequences cover indoor full-body motion, an outdoor dance sequence, and an upper-body subject performing …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

43 extracted references · 24 canonical work pages

  1. [1]

    Wan: Open and advanced large-scale video generative models.arXiv preprint arXiv:2503.20314,

    Alibaba Wan Team. Wan: Open and advanced large-scale video generative models.arXiv preprint arXiv:2503.20314,

    arXiv

  2. [2]

    Stable video diffusion: Scaling latent video diffusion models to large datasets.arXiv preprint arXiv:2311.15127, 2023

    Andreas Blattmann, Tim Dockhorn, Sumith Kulal, Daniel Mendelevitch, Maciej Kilian, Dominik Lorenz, Yam Levi, Zion English, Vikram V oleti, Adam Letts, et al. Stable video diffusion: Scaling latent video diffusion models to large datasets.arXiv preprint arXiv:2311.15127, 2023. 2

    arXiv 2023

  3. [3]

    Align your latents: High-resolution video synthesis with la- tent diffusion models

    Andreas Blattmann, Robin Rombach, Huan Ling, Tim Dock- horn, Seung Wook Kim, Sanja Fidler, and Karsten Kreis. Align your latents: High-resolution video synthesis with la- tent diffusion models. InCVPR, 2023. 2

    work page 2023

  4. [4]

    Emerg- ing properties in self-supervised vision transformers

    Mathilde Caron, Hugo Touvron, Ishan Misra, Hervé Jégou, Julien Mairal, Piotr Bojanowski, and Armand Joulin. Emerg- ing properties in self-supervised vision transformers. In ICCV, 2021. 8

    work page 2021

  5. [5]

    Diffusion forcing: Next-token prediction meets full-sequence diffu- sion

    Boyuan Chen, Diego Marti Monso, Yilun Du, Max Sim- chowitz, Russ Tedrake, and Vincent Sitzmann. Diffusion forcing: Next-token prediction meets full-sequence diffu- sion. InNeurIPS, 2024. 3

    work page 2024

  6. [6]

    Context forc- ing: Consistent autoregressive video generation with long context.arXiv preprint arXiv:2602.06028, 2026

    Shuo Chen, Cong Wei, Sun Sun, Ping Nie, Kai Zhou, Ge Zhang, Ming-Hsuan Yang, and Wenhu Chen. Context forc- ing: Consistent autoregressive video generation with long context.arXiv preprint arXiv:2602.06028, 2026. 3

    arXiv 2026

  7. [7]

    Wan-animate: Unified character anima- tion and replacement with holistic replication.arXiv preprint arXiv:2509.14055, 2025

    Gang Cheng, Xin Gao, Li Hu, Siqi Hu, Mingyang Huang, Chaonan Ji, et al. Wan-animate: Unified character anima- tion and replacement with holistic replication.arXiv preprint arXiv:2509.14055, 2025. 2, 3, 8

    arXiv 2025

  8. [8]

    Freeman, and Michael Rubinstein

    Ariel Ephrat, Inbar Mosseri, Oran Lang, Tali Dekel, Kevin Wilson, Avinatan Hassidim, William T. Freeman, and Michael Rubinstein. Looking to listen at the cocktail party: A speaker-independent audio-visual model for speech sepa- ration. InACM TOG, 2018. 7

    work page 2018

Show all 43 references
  1. [9]

    Usp: A unified sequence parallelism approach for long context generative ai.arXiv preprint arXiv:2405.07719, 2024

    Jiarui Fang and Shangchun Zhao. Usp: A unified sequence parallelism approach for long context generative ai.arXiv preprint arXiv:2405.07719, 2024. 2, 7

    2024 arXiv

  2. [10]

    Humandit: Pose-guided diffusion transformer for long- form human motion video generation.arXiv preprint arXiv:2502.04847, 2025

    Qijun Gan, Yi Ren, Chen Zhang, Zhenhui Ye, Pan Xie, Xiang Yin, Zehuan Yuan, Bingyue Peng, and Jianke Zhu. Humandit: Pose-guided diffusion transformer for long- form human motion video generation.arXiv preprint arXiv:2502.04847, 2025. 3

    2025 arXiv

  3. [11]

    Gans trained by a two time-scale update rule converge to a local nash equilib- rium

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilib- rium. InNeurIPS, 2017. 8

    2017

  4. [12]

    Denoising diffu- sion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffu- sion probabilistic models. InNeurIPS, 2020. 2

    2020

  5. [13]

    Video dif- fusion models.arXiv preprint arXiv:2204.03458, 2022

    Jonathan Ho, Tim Salimans, Alexey Gritsenko, William Chan, Mohammad Norouzi, and David J Fleet. Video dif- fusion models.arXiv preprint arXiv:2204.03458, 2022. 2

    2022 arXiv

  6. [14]

    Animate anyone: Consistent and controllable image-to-video synthesis for character animation.arXiv preprint arXiv:2311.17117, 2024

    Li Hu, Xin Gao, Peng Zhang, Ke Sun, Bang Zhang, and Liefeng Bo. Animate anyone: Consistent and controllable image-to-video synthesis for character animation.arXiv preprint arXiv:2311.17117, 2024. 2, 3

    2024 arXiv

  7. [15]

    Multianimate: Pose- guided image animation made extensible.arXiv preprint arXiv:2602.21581, 2026

    Yingcheng Hu, Haowen Gong, Chuanguang Yang, Zhulin An, Yongjun Xu, and Songhua Liu. Multianimate: Pose- guided image animation made extensible.arXiv preprint arXiv:2602.21581, 2026. 3

    2026 arXiv

  8. [16]

    Self forcing: Bridging the train-test gap in autoregressive video diffusion

    Xun Huang, Zhengqi Li, Guande He, Mingyuan Zhou, and Eli Shechtman. Self forcing: Bridging the train-test gap in autoregressive video diffusion. InNeurIPS, 2025. 2, 3, 4

    2025

  9. [17]

    Learning high fidelity depths of dressed humans by watching social media dance videos

    Yasamin Jafarian and Hyun Soo Park. Learning high fidelity depths of dressed humans by watching social media dance videos. InProceedings of the IEEE/CVF conference on com- puter vision and pattern recognition, pages 12753–12762,

  10. [18]

    Vace: All-in-one video creation and editing

    Zeyinzi Jiang, Zhen Han, Chaojie Mao, Jingfeng Zhang, Yulin Pan, and Yu Liu. Vace: All-in-one video creation and editing. InICCV, pages 17191–17202, 2025. 3

    2025

  11. [19]

    Rolling sink: Bridging limited-horizon training and open-ended testing in autoregressive video diffusion.arXiv preprint arXiv:2602.07775, 2026

    Haodong Li, Shaoteng Liu, Zhe Lin, and Manmohan Chan- draker. Rolling sink: Bridging limited-horizon training and open-ended testing in autoregressive video diffusion.arXiv preprint arXiv:2602.07775, 2026. 3

    2026 arXiv

  12. [20]

    Everanimate: Minute-scale human animation via latent flow restoration

    Wuyang Li, Yang Gao, Mariam Hassan, Lan Feng, Wentao Pan, Po-Chien Luan, and Alexandre Alahi. Everanimate: Minute-scale human animation via latent flow restoration. arXiv preprint arXiv:2605.15042, 2026. 2, 3, 8

    2026 arXiv

  13. [21]

    Stable video infinity: Infinite-length video generation with error recycling

    Wuyang Li, Wentao Pan, Po-Chien Luan, Yang Gao, and Alexandre Alahi. Stable video infinity: Infinite-length video generation with error recycling. InICLR, 2026. 3

    2026

  14. [22]

    Rolling forcing: Autoregressive long video diffusion in real time

    Kunhao Liu, Wenbo Hu, Jiale Xu, Ying Shan, and Shijian Lu. Rolling forcing: Autoregressive long video diffusion in real time. InICLR, 2026. 3

    2026

  15. [23]

    Scalable diffusion models with transformers

    William Peebles and Saining Xie. Scalable diffusion models with transformers. InICCV, 2023. 2

    2023

  16. [24]

    High-resolution image syn- thesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image syn- thesis with latent diffusion models. InCVPR, 2022. 2

    2022

  17. [25]

    One-to-all ani- mation: Alignment-free character animation and image pose transfer

    Shijun Shi, Jing Xu, Zhihang Li, Chunli Peng, Xiaoda Yang, Lijing Lu, Kai Hu, and Jiangning Zhang. One-to-all ani- mation: Alignment-free character animation and image pose transfer. InCVPR, pages 4011–4021, 2026. 2, 3, 8

    2026

  18. [26]

    First order motion model for image animation

    Aliaksandr Siarohin, Stephane Lathuiliere, Sergey Tulyakov, Elisa Ricci, and Nicu Sebe. First order motion model for image animation. InNeurIPS, 2019. 3

    2019

  19. [27]

    Videomae: Masked autoencoders are data-efficient learners for self-supervised video pre-training

    Zhan Tong, Yibing Song, Jue Wang, and Limin Wang. Videomae: Masked autoencoders are data-efficient learners for self-supervised video pre-training. InNeurIPS, 2022. 8

    2022

  20. [28]

    Musepose: A pose-driven image-to-video framework for virtual human generation.https : / / github.com/TMElyralab/MusePose, 2024

    Zhengyan Tong, Chao Li, Zhaokang Chen, Bin Wu, and Wenjiang Zhou. Musepose: A pose-driven image-to-video framework for virtual human generation.https : / / github.com/TMElyralab/MusePose, 2024. Open- source project. 3

    2024

  21. [29]

    Video-to- video synthesis

    Ting-Chun Wang, Ming-Yu Liu, Jun-Yan Zhu, Guilin Liu, Andrew Tao, Jan Kautz, and Bryan Catanzaro. Video-to- video synthesis. InNeurIPS, 2018. 3

    2018

  22. [30]

    Unianimate-dit: Human image animation with large-scale video diffusion transformer.arXiv preprint arXiv:2504.11289, 2025

    Xiang Wang, Shiwei Zhang, Longxiang Tang, Yingya Zhang, Changxin Gao, Yuehuan Wang, and Nong Sang. Unianimate-dit: Human image animation with large-scale video diffusion transformer.arXiv preprint arXiv:2504.11289, 2025. 2, 3, 8

    2025

  23. [31]

    Humanvid: Demystifying train- ing data for camera-controllable human image animation

    Zhenzhi Wang, Yixuan Li, Yanhong Zeng, Youqing Fang, Yuwei Guo, Wenran Liu, Jing Tan, Kai Chen, Tianfan Xue, Bo Dai, and Dahua Lin. Humanvid: Demystifying train- ing data for camera-controllable human image animation. In NeurIPS, 2024. 7

    2024

  24. [32]

    A learning algorithm for continually running fully recurrent neural networks.Neu- ral Computation, 1(2):270–280, 1989

    Ronald J Williams and David Zipser. A learning algorithm for continually running fully recurrent neural networks.Neu- ral Computation, 1(2):270–280, 1989. 3

    1989

  25. [33]

    Q-Align: Teaching LMMs for visual scoring via discrete text-defined levels

    Haoning Wu, Zicheng Zhang, Weixia Zhang, Chaofeng Chen, Liang Liao, Chunyi Li, Yixuan Gao, Annan Wang, Erli Zhang, Wenxiu Sun, Qiong Yan, Xiongkuo Min, Guang- tao Zhai, and Weisi Lin. Q-Align: Teaching LMMs for visual scoring via discrete text-defined levels. InProceedings of ...

    2024

  26. [34]

    Efficient streaming language models with attention sinks

    Guangxuan Xiao, Yuandong Tian, Beidi Chen, Song Han, and Mike Lewis. Efficient streaming language models with attention sinks. InICLR, 2024. 4

    2024

  27. [35]

    Magicanimate: Temporally consistent human im- age animation using diffusion model.arXiv preprint arXiv:2311.16498, 2024

    Zhongcong Xu, Jianfeng Zhang, Jun Hao Liew, Hanshu Yan, Jia-Wei Liu, Chenxu Zhang, Jiashi Feng, and Mike Zheng Shou. Magicanimate: Temporally consistent human im- age animation using diffusion model.arXiv preprint arXiv:2311.16498, 2024. 2, 3

    2024 arXiv

  28. [36]

    Scail: Towards studio-grade character animation via in-context learning of 3d-consistent pose representations

    Wenhao Yan, Sheng Ye, Zhuoyi Yang, Jiayan Teng, Zhen- Hui Dong, Kairui Wen, Xiaotao Gu, Yong-Jin Liu, and Jie Tang. Scail: Towards studio-grade character animation via in-context learning of 3d-consistent pose representations. arXiv preprint arXiv:2512.05905, 2025. 2, 3, 8

    2025

  29. [37]

    Freeman, and Taesung Park

    Tianwei Yin, Michaël Gharbi, Richard Zhang, Eli Shecht- man, Frédo Durand, William T. Freeman, and Taesung Park. One-step diffusion with distribution matching distillation. In CVPR, pages 6613–6623, 2024. 3

    2024

  30. [38]

    From slow bidirectional to fast autoregressive video diffusion mod- els

    Tianwei Yin, Qiang Zhang, Richard Zhang, William T Free- man, Frédo Durand, Eli Shechtman, and Xun Huang. From slow bidirectional to fast autoregressive video diffusion mod- els. InCVPR, 2025. 3

    2025

  31. [39]

    Steadydancer: Harmonized and coherent human image animation with first-frame preserva- tion.arXiv preprint arXiv:2511.19320, 2025

    Jiaming Zhang, Shengming Cao, Rui Li, Xiaotong Zhao, Yu- tao Cui, Xinglin Hou, Gangshan Wu, Haolan Chen, Xu Yu, Limin Wang, and Kai Ma. Steadydancer: Harmonized and coherent human image animation with first-frame preserva- tion.arXiv preprint arXiv:2511.19320, 2025. 3, 7

    2025 arXiv

  32. [40]

    UniPC: A unified predictor-corrector framework for fast sampling of diffusion models

    Wenliang Zhao, Lujia Bai, Yongming Rao, Jie Zhou, and Jiwen Lu. UniPC: A unified predictor-corrector framework for fast sampling of diffusion models. InNeurIPS, 2023. 8

    2023

  33. [41]

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

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

    arXiv

  34. [42]

    Champ: Controllable and consistent human image ani- mation with 3d parametric guidance

    Shenhao Zhu, Junming Leo Chen, Zuozhuo Dai, Zilong Dong, Yinghui Xu, Xun Cao, Yao Yao, Hao Zhu, and Siyu Zhu. Champ: Controllable and consistent human image ani- mation with 3d parametric guidance. InECCV, 2024. 3 LiveAnimate: Stable Long-Form Streaming Human Animation in Real...

    2024

  35. [43]

    Additional Qualitative Results We provide seven additional three-minute examples to complement the qualitative comparisons in Sec. 4. Fig- ures 8 and 9 cover full-body and upper-body animation un- der diverse identities, clothing, backgrounds, camera fram- ing, and motion patt...

Pith tools

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