Vorch-IR: Long-Form Unified Multimodal Identity Replacement Video Generation

Gang Yue; Jingjing Chen; Lin Ma; Xiaoyu Chen; Xin Ma; Yang Ding; Yaohui Wang; Yaole Wang

REVIEW 4 major objections 6 minor 15 references

Vorch-IR: Long-Form Unified Multimodal Identity Replacement Video Generation

T0 review · 4 major / 6 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read One model replaces one or two people in a video, and optionally the background, from reference images and a text instruction, with no masks or pose inputs.

desk verdict Solid single-person results, but the paper's distinctive claims—pose-decoupled references, dual-person and background replacement, long-video stability—are not backed by the reported experiments. read the letter →

arxiv 2608.05648 v1 pith:GM74QPPW submitted 2026-08-06 cs.CV

Yaole Wang , Xiaoyu Chen , Xin Ma , Yang Ding , Gang Yue , Jingjing Chen , Lin Ma , Yaohui Wang This is my paper · ORCID

classification cs.CV

keywords videoidentityreplacementmulti-personeditingbackgrounddiffusiontransformerreference-conditionedgenerationlanguage-guidedcorrespondencelong-formautomaticdataconstruction

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

Vorch-IR tries to show that identity replacement in video does not need task-specific masks, pose maps, or spatially aligned references: a single diffusion model can replace one or two subjects, and optionally the background, from a driving video plus one or more reference images and a text instruction that says which reference goes where. The point of unifying the four settings is practical: users get one interface for single- and dual-person swaps, with or without scene changes, and the model shares its parameters across all four tasks. The paper also claims that the scarcity of paired training data can be bypassed with an automatic pipeline that edits the first frame, propagates the edit through a motion-guided animation model, filters failures, and extracts references, and that a temporal overlapping inference scheme extends the short-clip model to minute-long generation without autoregressive continuation. If right, the result is a single multimodal editing system, not a collection of specialized animation tools.

What carries the argument

The load-bearing object is the unified visual token stream: VAE latents of the driving video, each indexed reference, and the noisy target are concatenated and processed by a shared Diffusion Transformer through self-attention, with task-ID embeddings marking each input slot and giving driving and reference tokens timestep zero while only target tokens are noised. Alongside this dense path, a Gemma vision-language encoder processes the instruction, sampled driving frames, and references to produce a context vector that the target tokens query through cross-attention, so reference-to-subject assignment is learned as language-grounded semantics rather than geometry. For minute-long outputs, the adapted Bidirectional Latent Fusion divides the full target latent into overlapping temporal windows, fuses window predictions with a linear or cosine cross-fade, and performs a single global Euler step per timestep, which is what prevents autoregressive drift.

What would settle it

Run the paper's data pipeline on a sample of source clips and measure identity similarity between the reference image and the propagated first frame with a feature metric such as DINOv2; if the pipeline's synthetic targets show low or unreliable identity similarity, Vorch-IR's reported identity-preservation scores cannot be taken at face value. A direct check is to compare Vorch-IR's output on pipeline samples where the animation model visibly fails against samples where it succeeds: if output quality tracks the animation model's failures, the synthetic ground truth is carrying the result.

Watch

Extended reading notes

Core claim

The central claim is that one diffusion transformer, Vorch-IR, performs single-person replacement, dual-person replacement, and both with background replacement by jointly conditioning on a driving video, indexed reference images, and a textual instruction, with no masks or pose inputs at inference. Reference images are spatially decoupled from the driving video: their order fixes the slot and the instruction fixes the role, so 'Reference 1 replaces the person on the left' is handled as a semantic relation rather than through aligned coordinates or region localization. Dense motion and appearance are fused by concatenating driving-video tokens, reference tokens, and the noisy target into one self-attention stream with task-ID embeddings, while a vision-language encoder supplies cross-attention context that binds each reference to its target. All four editing settings are trained jointly with one flow-matching loss on pairs produced by an automatic data-construction pipeline. For long videos, a temporal overlapping inference strategy adapted from bidirectional latent fusion denoises the full target latent through overlapping windows and applies one global update per step, avoiding autoregressive continuation and keeping the whole sequence on a single diffusion trajectory.

Load-bearing premise

The training labels come from an automatic pipeline that edits a first frame, animates it with a motion-guided model, and filters with a vision-language model, so the whole method inherits the animation model's errors: if that model does not truly preserve the reference identity and source motion, Vorch-IR learns to reproduce those errors and the identity metrics measure agreement with synthetic labels rather than real replacement quality.

Editorial extensions

If this is right

The same model parameters handle all four settings, so single-person and dual-person swaps, with or without background changes, share one input interface and one training objective.
Since no masks or pose inputs are needed, the pipeline can be wrapped in a generic image-text-video interface where the user only provides reference images and a sentence.
Long, minute-scale videos are generated on one global denoising trajectory via overlapping window fusion, avoiding the identity drift and error accumulation of autoregressive continuation.
The automatic data-construction pipeline removes the main obstacle to scaling paired identity-replacement data, since it synthesizes training tuples for all four settings from raw single- and dual-person clips.
Background replacement is treated as just another indexed reference, so scene editing and identity editing are unified in the same conditioning mechanism.

Reading between the lines

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

If the synthetic training pairs are faithful, the indexed-reference-plus-instruction recipe should transfer to other reference-driven edits such as object replacement, style transfer, or relighting, because the interface does not depend on faces or poses.
The overlapping-window fusion idea is generic: any frame-aligned video editor that has a complete driving condition could denoise the full latent through overlapping windows and avoid autoregressive drift, so it is testable on other editing tasks.
The paper's direct comparisons cover only the shared single-person setting; the dual-person and background claims rest on qualitative demonstrations, so a multi-person benchmark against mask-guided baselines would be needed to know how language-based reference binding compares in practice.
One practical limit to watch is ambiguous instructions or visually similar references, since assignment is learned semantically rather than by geometry.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

First thing to know: this paper is a solid single-person identity replacement system with a plausible unified interface, but its headline claims about pose-decoupled references and multi-person/background replacement are not actually demonstrated in the paper. The single-person numbers are competitive, the human eval is reported in detail, and the cross-dataset XDance result is a good idea — but the table carrying the DINOv2/CLIP-I numbers is missing ('Tab.??'), and the dual-person and background settings get only qualitative results on a project page.

What's new: the indexed reference interface plus instruction-based binding is a real step beyond mask- or pose-conditioned methods, and the synthetic data pipeline that generates paired tuples for all four settings is a practical contribution. The overlapping latent fusion for long video is a sensible adaptation of BLF, though it's only validated via project-page videos.

Soft spots, in order of severity. First, the stress-test concern is correct. Every training reference is cropped from the edited first frame, which is pose-aligned with the driving video. The paper claims references 'need not match the pose, layout, or spatial configuration' but provides no training example of a misaligned reference and no evaluation that varies reference pose. Self-attention can learn pose-aligned appearance transfer; the VLM cross-attention is never trained on decoupled pairs. So that central interface claim is currently unsubstantiated. Second, the synthetic targets come from the same pipeline used to build the benchmark, which creates a circularity problem: identity-preservation metrics may partly measure the model's ability to reconstruct the animation model's output, not true replacement quality. Third, the dual-person and background settings — the paper's signature — have no quantitative evaluation. Fourth, no artifacts, no error bars, and a missing table.

None of this is fatal to the underlying idea. The single-person results against strong baselines are credible, and the architecture is clearly explained. But the paper as written overclaims. It deserves a serious referee, and the revision should add: a pose-decoupling experiment (e.g., reference images from different viewpoints/poses), quantitative dual-person and background results, error bars, and either code/data or a much more detailed project page. I'd send it to review, but with the expectation of major revision.

Referee Report

4 major / 6 minor

Summary. The paper proposes Vorch-IR, a unified video-generation model for single- and dual-person identity replacement with optional background replacement. Built on the LTX2 diffusion transformer, the model conditions on a driving video, indexed reference images, and a textual instruction, fusing dense visual conditions via self-attention and instruction-grounded semantics via a frozen Gemma vision-language model through cross-attention. The authors introduce an automatic data-construction pipeline that edits the first frame of a source clip with an off-the-shelf editor, propagates the edit with a motion-guided animation model, filters failures, and crops references from the synthesized first frame. A shared flow-matching objective trains all four settings, and a temporal overlapping-inference strategy (adapted from BLF) extends the short-clip model to minute-long outputs. Quantitative evaluation is limited to the single-person setting on an internal benchmark and XDance, with GSB pairwise human evaluation; dual-person, background-replacement, and long-form results are only on a project page. The paper makes central claims about spatial decoupling of references and unified multi-person editing that require additional evidence.

Significance. If the central claims were fully substantiated, Vorch-IR would be a notable step toward a single model that handles multi-person identity replacement and background editing through a unified, mask-free, pose-free interface. The architecture is plausible: self-attention over a concatenated token stream for dense appearance/motion transfer, VLM cross-attention for reference-to-target binding, and a single flow-matching objective for all tasks. The temporal overlapping-inference adaptation is principled and could be useful for long-form editing. The automatic data pipeline, if it produced true ground truth, would address the data-scarcity problem. However, the evidence provided does not yet establish the central claims: the training data generation has a circularity problem (references are cropped from the synthetic target's first frame), the spatial-decoupling interface is never tested with misaligned references, the signature multi-person and background settings lack quantitative evaluation, and a key XDance identity-similarity table is missing. These gaps make the current contribution more of a systems proposal with promising initial results than a fully validated method.

major comments (4)

[Sec. 3.2 and Sec. 3.1] The claimed interface that references 'need not match the pose, layout, or spatial configuration of the driving video' (Sec. 1, Sec. 3.1) is not realized in the training distribution: Sec. 3.2 states that the pipeline edits the first frame 'while approximately preserving subject locations and coarse body poses,' then 'crops the edited subjects from the synthesized first frame to form identity references.' Every identity reference in every training tuple is therefore aligned with the driving video's first-frame layout, and no geometric augmentation or pose-decoupling ablation is reported. Self-attention (Eqs. 3-5) can learn to copy appearance from the aligned reference tokens to the corresponding target tokens at the first frame, so the model's demonstrated ability on the aligned test set does not establish transfer to arbitrarily posed references. At minimum, the authors should evaluate on references with altered pose/layout and add training augmentations (random crop, rotate, flip on reference images) and report whether performance degrades.
[Sec. 4.2] All quantitative comparisons (Tables 1-3, Fig. 6) are restricted to the single-person replacement setting. The tasks highlighted in the title and abstract — dual-person replacement, background replacement, and minute-long generation — are presented only as qualitative 'project page' results, with no metrics or in-paper examples. The abstract's claim of 'strong identity preservation, motion fidelity, and temporal coherence across diverse scenarios' is therefore only partially supported. The authors should provide quantitative evaluations (e.g., subject similarity, background consistency, temporal coherence) for dual-person and background-replacement settings, and at least one reproducible long-video evaluation (e.g., length, identity drift over time) instead of deferring to a project page.
[Sec. 4.2, XDance paragraph] The text in the XDance paragraph reports DINOv2 (0.6040), CLIP-I (0.7953), and DWPose (0.0761) values but refers to a table 'Tab.??' that does not appear in the manuscript. Without this table, the cross-dataset identity-preservation claim cannot be verified. In addition, the construction of 'our benchmark' is not described; if it is generated by the same automatic pipeline as the training data, the reported identity-similarity scores could reflect self-consistency with synthetic targets rather than genuine replacement fidelity. The authors should include the missing table and provide a clear description of benchmark construction (including whether references for the test set are derived from the target video's first frame).
[Sec. 3.2, Sec. 4.2] The training targets are produced by the same pipeline that yields the references: an image editor modifies the first frame, an animation model propagates it, and the same synthesized appearance is cropped back out as the reference. If the animation model does not perfectly preserve identity and motion, the network is trained to reproduce the animation model's output, and the reported metrics (e.g., subject consistency 0.9096 in Table 1) may be measuring reconstruction of the synthetic label rather than identity transfer to a new subject. The authors should acknowledge this limitation explicitly, validate the pipeline on a small set of human-verified real edit pairs, or show that metrics are insensitive to the choice of animation model.

minor comments (6)

[Sec. 4.2, Fig. 6] The GSB human evaluation reports percentages without the number of evaluators, the number of evaluated pairs, or confidence intervals, making the 65.4% vs 23.1% differences hard to interpret; these quantities should be reported.
[Sec. 3.4, Eq. (11)] The linear cross-fade in Eq. (11) uses alpha_k = k/(o-1); when o=1 the denominator is zero and this case is not defined. The implementation details should state the handling of the no-overlap case.
[Sec. 4.1] The training set of 15,571 clips totaling about 13 hours is on the small side for training a video diffusion model; the paper should discuss whether this is sufficient and should report the filtering rate of the data-construction pipeline.
[Sec. 4.1, Tables 1-2] The paper does not state how many test clips are used in 'our benchmark' or how the benchmark was split from the training set; dataset statistics and evaluation protocol should be added.
[Sec. 3.2] The paper should identify the specific off-the-shelf image editor and motion-guided animation model used in the data pipeline, as well as the VLM filtering criteria, for reproducibility.
[Figure 1] Figure 1 contains garbled text in the header, likely a rendering issue; the figures should be cleaned for the final version.

Circularity Check

1 steps flagged · score 4.0 of 10

The training pipeline crops identity references from the synthesized target's first frame, so the headline claim that references need not match pose or layout is never trained or tested; identity metrics partly measure self-consistency with synthetic labels.

self definitional [Sec. 3.2 (Data Construction Pipeline), 'The edited first frame and original clip ...' through '... form the final training tuple.']
"The edited first frame and original clip are then passed to a motion-guided video animation model, which propagates the new identities and scene along the source motion trajectories. ... Finally, we crop the edited subjects from the synthesized first frame to form identity references."

The identity reference R is defined as a crop of the first frame of the synthesized target video Y, while Y itself is produced by animating that same edited first frame. Every training example therefore has a reference that is pose/layout-aligned with the target by construction, and the reference appearance is exactly the appearance inserted into the ground-truth first frame. The Sec. 3.1 claim that references 'need not match the pose, layout, or spatial configuration of the driving video' is thus never exercised during training, and the 'our benchmark' evaluation built by the same pipeline inherits the same alignment.

full rationale

Vorch-IR's core architecture is not circular: the model fine-tunes an external LTX2 checkpoint, uses a frozen Gemma vision-language encoder for cross-attention, adopts a standard flow-matching objective, and adapts an external BLF-style inference method. No load-bearing argument reduces to a self-citation by the authors. The main circularity lies in the data-construction and evaluation loop of Sec. 3.2 and Sec. 4.2: identity references are cropped from the synthesized first frame, and the ground-truth video is the animation of that same edited first frame. Consequently, (i) all training references are pose/layout-aligned with the corresponding target, so the central interface claim that references need not match the driving video's pose or layout is untested by the training distribution; (ii) identity-preservation metrics on 'our benchmark', built by the same pipeline, largely measure self-consistency with synthetic labels; and (iii) the only external benchmark, XDance, leaves its identity-similarity table as 'Tab.??', so the claimed cross-distribution identity generalization is not actually reported. This warrants a moderate circularity score: the architecture and training objective remain independently grounded, but the headline pose-decoupling capability and the identity numbers are partially forced by the construction of the training targets.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The paper introduces no new physical entities or external falsifiable handles. Its machinery consists of internal architectural components (a task-ID embedding table, indexed reference slots, VLM cross-attention) that do not constitute invented entities in the sense of this ledger. The main audit findings are the free parameters of the inference and data pipelines, and the domain assumptions about the reliability of the synthetic targets and the VLM binding.

free parameters (4)

temporal window length and overlap (8s window, 2s overlap) = 8s window, 2s overlap (stride 6s)
Chosen by hand for long-video inference; affects quality and drift, not derived from data.
cross-fade weighting schedule = linear cross-fade (Eq. 11), cosine also supported
Hyperparameter choice for overlap fusion; other variants are noted but not used in the reported results.
training checkpoint (2,000 optimizer updates) = 2,000 updates
Stopping criterion chosen without a described validation protocol; affects all reported numbers.
VLM filtering criteria for data pipeline = unspecified thresholds for malformed bodies, identity mismatch, etc.
Hand-defined filter categories; rejection rates are not reported, so the sample selection is not auditable.

assumptions (5)

domain assumption Synthetic targets generated by an off-the-shelf motion-guided animation model accurately represent the identity and motion of the reference and driving video.
Sec. 3.2; if the animation model mismatches identity or motion, the training targets are mislabeled.
domain assumption The VLM context can map an indexed reference to the intended subject/background based on the instruction alone, without geometric alignment.
Sec. 3.1; the entire binding mechanism relies on Gemma's semantic understanding.
standard math Driving and target latents have identical dimensions so the independently constructed RoPE grids align with the same temporal origin.
Sec. 3.1, Eqs. 3-4; needed for frame-aligned self-attention.
standard math Flow-matching objective (Eq. 10) is a valid training signal for conditional video generation.
Standard objective, cited (Lipman et al., 2023).
domain assumption Public datasets plus internal collections form a broad enough training distribution for the four editing settings.
Sec. 4.1; no dataset statistics beyond counts and durations are provided.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Vorch-IR: Long-Form Unified Multimodal Identity Replacement Video Generation." pith.science (2026). https://pith.science/paper/GM74QPPW

@misc{pith2026260805648,
  author       = {Pith},
  title        = {Pith review of: Vorch-IR: Long-Form Unified Multimodal Identity Replacement Video Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GM74QPPW}},
  note         = {Machine review of arXiv:2608.05648}
}

read the original abstract

Video identity replacement seeks to transfer the identities of one or more subjects while preserving the motion, expressions, and temporal structure of a driving video. Existing methods largely target single-person settings and often require task-specific structural controls, such as masks or pose representations, limiting their flexibility in general multimodal editing systems. Progress on multi-person replacement is further constrained by the scarcity of paired training data. We present Vorch-IR, a unified framework that supports single- and dual-person identity replacement, with optional background replacement, in a single model. Built on LTX2, Vorch-IR jointly conditions on a driving video, indexed reference images, and a textual editing instruction. The reference images need not match the pose, layout, or spatial configuration of the driving video: their roles as subject or background references are specified through the instruction. Dense visual conditions are fused through self-attention, while a vision-language context establishes semantic correspondence through cross-attention. We further develop an automatic data construction pipeline that synthesizes paired supervision for all four editing settings. Experiments using automatic metrics and pairwise human evaluation demonstrate strong identity preservation, motion fidelity, and temporal coherence across diverse scenarios. A temporal overlapping inference strategy additionally extends the short-clip model to minute-long generation without autoregressive continuation.

Figures

Figures reproduced from arXiv: 2608.05648 by the authors.

**Figure 1.** Representative results of Vorch-IR. A single model performs single- and dual-person identity replacement, with or without background editing, while preserving the motion and temporal structure of the driving video. The same model can be applied to minute-long sequences through temporal overlapping inference. Abstract Video identity replacement seeks to transfer the identities of one or more subjects while preservin… view at source ↗

**Figure 2.** Overview of Vorch-IR. The driving video, indexed reference images, and noisy target latent form a unified visual token stream with task-ID embeddings. A Gemma vision-language encoder jointly processes the instruction, sampled driving frames, and references to produce cross-attention context. A shared Diffusion Transformer combines frame-aligned motion guidance with instruction-grounded reference appearance. specifie… view at source ↗

**Figure 3.** Overview of four identity replacement settings. A single model performs single-person and dual-person replacement, with optional background replacement. The textual instruction associates each indexed reference image with a subject or the background in the driving video. Here, the textual markers denote the ordering of image placeholders in the task template; no special background marker is required. A vision-langu… view at source ↗

Figures from the paper (3 more)

**Figure 4.** Figure 4: Automatic data construction. We edit the identities and, when required, the background of the first frame; propagate the edited appearance through the source motion; filter failed generations; and extract identity and background references to form paired training tuple…

**Figure 5.** Figure 5: Single-person replacement comparison. Generated first frames from Wan-Animate, SCAIL-2, and Vorch-IR. Vorch-IR closely preserves the reference appearance while following the pose and composition of the driving video. not expose a common interface for dual-person and ba…

**Figure 6.** Figure 6: Good–Same–Bad (GSB) pairwise human evaluation. For each dimension and baseline, Good indicates that evaluators prefer Vorch-IR, Same indicates comparable perceptual quality, and Bad indicates a preference for the competing method. The numbers are percentages, and each …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

15 extracted references · 13 linked inside Pith

[1]

Diffusion forcing: Next-token prediction meets full-sequence diffusion.arXiv preprint arXiv:2407.01392, 2024a

Boyuan Chen, Diego Marti Monso, Yilun Du, Max Simchowitz, Russ Tedrake, and Vincent Sitzmann. Diffusion forcing: Next-token prediction meets full-sequence diffusion.arXiv preprint arXiv:2407.01392, 2024a. Xinyuan Chen, Yaohui Wang, Lingjun Zhang, Shaobin Zhuang, Xin Ma, Jiashuo Yu, Yali Wang, Dahua Lin, Yu Qiao, and Ziwei Liu. Seine: Short-to-long video d...

arXiv
[3]

Seedance 1.0: Exploring the boundaries of video generation models.arXiv preprint arXiv:2506.09113,

Yu Gao, Haoyuan Guo, Tuyen Hoang, Weilin Huang, Lu Jiang, Fangyuan Kong, Huixia Li, Jiashi Li, Liang Li, Xiaojie Li, Xunsong Li, Yifu Li, Shanchuan Lin, Zhijie Lin, Jiawei Liu, Shu Liu, et al. Seedance 1.0: Exploring the boundaries of video generation models.arXiv preprint arXiv:2506.09113,

arXiv
[4]

Ltx-2: Efficient joint audio-visual foundation model.arXiv preprint arXiv:2601.03233,

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,

arXiv
[6]

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

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,

arXiv
[7]

Anyv2v: A tuning-free framework for any video-to-video editing tasks.arXiv preprint arXiv:2403.14468,

Max Ku, Cong Wei, Weiming Ren, Harry Yang, and Wenhu Chen. Anyv2v: A tuning-free framework for any video-to-video editing tasks.arXiv preprint arXiv:2403.14468,

arXiv
[9]

One-to-all animation: Alignment-free character animation and image pose transfer.arXiv preprint arXiv:2511.22940,

Shijun Shi, Jing Xu, Zhihang Li, Chunli Peng, Xiaoda Yang, Lijing Lu, Kai Hu, and Jiangning Zhang. One-to-all animation: Alignment-free character animation and image pose transfer.arXiv preprint arXiv:2511.22940,

arXiv
[10]

Hunyuanvideo: A systematic framework for large video generative models.arXiv preprint arXiv:2412.03603,

Tencent Hunyuan Team. Hunyuanvideo: A systematic framework for large video generative models.arXiv preprint arXiv:2412.03603,

arXiv
[12]

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

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,

arXiv

Show all 15 references

[13]

Scail-2: Unifying controlled character animation with end-to-end in-context conditioning.arXiv preprint arXiv:2606.10804, 2026a

Wenhao Yan, Fengjia Guo, Zhuoyi Yang, and Jie Tang. Scail-2: Unifying controlled character animation with end-to-end in-context conditioning.arXiv preprint arXiv:2606.10804, 2026a. Wenhao Yan, Sheng Ye, Zhuoyi Yang, Jiayan Teng, ZhenHui Dong, Kairui Wen, Xiaotao Gu, Yong-Jin L...

arXiv
[14]

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

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

arXiv
[15]

Ni, and Heung-Yeung Shum

Deyu Zhou, Quan Sun, Yuang Peng, Kun Yan, Runpei Dong, Duomin Wang, Zheng Ge, Nan Duan, Xiangyu Zhang, Lionel M. Ni, and Heung-Yeung Shum. Taming teacher forcing for masked autoregressive video generation.arXiv preprint arXiv:2501.12389,

arXiv
[2023]

Dreamactor-m2: Universal character image animation via spatiotemporal in-context learning.arXiv preprint arXiv:2601.21716,

Mingshuang Luo, Shuang Liang, Zhengkun Rong, Yuxuan Luo, Tianshu Hu, Ruibing Hou, Hong Chang, Yong Li, Yuan Zhang, and Mingyuan Gao. Dreamactor-m2: Universal character image animation via spatiotemporal in-context learning.arXiv preprint arXiv:2601.21716,
[2024]

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

13 Vorch-IR: Long-Form Unified Multimodal Identity Replacement Video Generation Wan Team. Wan: Open and advanced large-scale video generative models.arXiv preprint arXiv:2503.20314,

arXiv
[2025]

Skyreels-audio: Omni audio-conditioned talking portraits in video diffusion transformers.arXiv preprint arXiv:2506.00830,

12 Vorch-IR: Long-Form Unified Multimodal Identity Replacement Video Generation Zhengcong Fei, Hao Jiang, Di Qiu, Baoxuan Gu, Youqiang Zhang, Jiahua Wang, Jialin Bai, Debang Li, Mingyuan Fan, Guibin Chen, and Yahui Zhou. Skyreels-audio: Omni audio-conditioned talking portraits...

arXiv
[2026]

Animate anyone 2: High-fidelity character image animation with environment affordance

Li Hu, Guangyuan Wang, Zhen Shen, Xin Gao, Dechao Meng, Lian Zhuo, Peng Zhang, Bang Zhang, and Liefeng Bo. Animate anyone 2: High-fidelity character image animation with environment affordance. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 102...

arXiv

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.

[1] [1]

Diffusion forcing: Next-token prediction meets full-sequence diffusion.arXiv preprint arXiv:2407.01392, 2024a

Boyuan Chen, Diego Marti Monso, Yilun Du, Max Simchowitz, Russ Tedrake, and Vincent Sitzmann. Diffusion forcing: Next-token prediction meets full-sequence diffusion.arXiv preprint arXiv:2407.01392, 2024a. Xinyuan Chen, Yaohui Wang, Lingjun Zhang, Shaobin Zhuang, Xin Ma, Jiashuo Yu, Yali Wang, Dahua Lin, Yu Qiao, and Ziwei Liu. Seine: Short-to-long video d...

arXiv

[2] [3]

Seedance 1.0: Exploring the boundaries of video generation models.arXiv preprint arXiv:2506.09113,

Yu Gao, Haoyuan Guo, Tuyen Hoang, Weilin Huang, Lu Jiang, Fangyuan Kong, Huixia Li, Jiashi Li, Liang Li, Xiaojie Li, Xunsong Li, Yifu Li, Shanchuan Lin, Zhijie Lin, Jiawei Liu, Shu Liu, et al. Seedance 1.0: Exploring the boundaries of video generation models.arXiv preprint arXiv:2506.09113,

arXiv

[3] [4]

Ltx-2: Efficient joint audio-visual foundation model.arXiv preprint arXiv:2601.03233,

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,

arXiv

[4] [6]

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

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,

arXiv

[5] [7]

Anyv2v: A tuning-free framework for any video-to-video editing tasks.arXiv preprint arXiv:2403.14468,

Max Ku, Cong Wei, Weiming Ren, Harry Yang, and Wenhu Chen. Anyv2v: A tuning-free framework for any video-to-video editing tasks.arXiv preprint arXiv:2403.14468,

arXiv

[6] [9]

One-to-all animation: Alignment-free character animation and image pose transfer.arXiv preprint arXiv:2511.22940,

Shijun Shi, Jing Xu, Zhihang Li, Chunli Peng, Xiaoda Yang, Lijing Lu, Kai Hu, and Jiangning Zhang. One-to-all animation: Alignment-free character animation and image pose transfer.arXiv preprint arXiv:2511.22940,

arXiv

[7] [10]

Hunyuanvideo: A systematic framework for large video generative models.arXiv preprint arXiv:2412.03603,

Tencent Hunyuan Team. Hunyuanvideo: A systematic framework for large video generative models.arXiv preprint arXiv:2412.03603,

arXiv

[8] [12]

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

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,

arXiv