RDPO: Real Data Preference Optimization for Physics Consistency Video Generation

Anxiang Zeng; Chaoyue Wang; Hao Li; Hou Peng; Wenxu Qian; Zhiyu Tan

REVIEW 4 major objections 5 minor 1 cited by

RDPO: Real Data Preference Optimization for Physics Consistency Video Generation

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read RDPO shows that preference optimization for video physics can run on real-video priors alone, without human annotation.

desk verdict Clever annotation-free preference construction, but the paper doesn't yet prove it's preference learning rather than real-video distillation. read the letter →

arxiv 2506.18655 v1 pith:O4PYEWWY submitted 2025-06-23 cs.CV

Wenxu Qian , Chaoyue Wang , Hou Peng , Zhiyu Tan , Hao Li , Anxiang Zeng This is my paper

classification cs.CV

keywords videogenerationphysicsconsistencypreferenceoptimizationDPOannotation-freediffusionmodelsrectifiedflowrealdataprior

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 tries to establish that a video generation model can be taught real-world physics without any human annotation, using the model's own relationship to real videos. The claim is that by partially noising real footage and then reverse-sampling it with the pretrained generator, one can automatically produce a 'preferred' video that is statistically more physically correct than a video generated from pure noise. Training the generator with a direct preference optimization objective on these self-built pairs improves physical realism and overall video quality. If true, it removes the main bottleneck of preference-based post-training for video: costly human labels or brittle reward models.

What carries the argument

The load-bearing mechanism is reverse sampling from a real video's noised latent (an idea borrowed from image-to-image editing), steered by rejection sampling: among K noised versions of the real latent at timestep t, the one closest in L2 distance to the model's own sample at that timestep is chosen for denoising. This keeps the 'preferred' video aligned with the model's manifold while injecting real-world dynamics. The Flow-DPO objective (Equation 3) then raises the likelihood of x_preferred relative to x_model, and a progressive curriculum—starting from heavy noise and decreasing reverse steps across iterations—pulls the model toward the real-data distribution.

What would settle it

Take the trained RDPO model and feed it preference pairs where x_preferred is replaced by the original real video (or a fully denoised copy of it). If the PhysicsIQ gain disappears or is unchanged compared to plain SFT on real videos, the preference signal carries no information beyond the real-video prior itself.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that preference pairs for physics consistency can be manufactured from real videos at no annotation cost. The method takes a real video, adds noise to a chosen intermediate timestep, and reverse-samples that noisy latent to get x_preferred, while a normal generation from full noise yields x_model. Because the reverse-sampled output retains dynamic structure from the real footage, it is usually closer to physical plausibility; the pair (x_preferred, x_model) then serves as a synthetic DPO example. Applied with a coarse-to-fine schedule of reverse steps and combined with SFT, this raises LTX-2B's PhysicsIQ from 24.40 to 25.66 over three iterations and matches or exceeds human-annotated DPO on PhysicsIQ and Total Score.

Load-bearing premise

The claim rests on the assumption that a video produced by partially denoising a real video's latent is consistently more physically correct than a video generated from full noise, and that this preference survives text-conditioned generation; if the 'preferred' video is essentially just a reconstruction of the real footage, the DPO signal collapses into supervised fine-tuning.

Editorial extensions

If this is right

If correct, any text-to-video diffusion model can be post-trained for physics consistency without building reward models or collecting human preferences, using only a corpus of real videos.
The progressive reverse-step curriculum suggests a general recipe: start with high noise to teach global dynamics, then decrease noise for fine-grained physical detail, interleaving earlier phases to avoid catastrophic forgetting.
RDPO's auto-built pairs are complementary to human-labeled DPO: mixing them gives the best scores, implying the two supervision sources carry partially independent signal.
The approach may transfer to other attributes beyond physics—whatever distributional structure real videos carry—provided the dataset is curated for that attribute.

Reading between the lines

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

A likely implicit limit: because Algorithm 1 does not condition the reverse sampling on the text prompt, x_preferred may not correspond to the same semantic content as x_model's prompt; if the pair does not share the prompt c, the DPO update could be teaching prompt-agnostic imitation rather than physics per se.
One could test whether the gains scale with diversity of real videos: if the physics prior is purely distributional, a wider set of real scenes should yield larger PhysicsIQ improvements; if not, the method may be memorizing specific dynamics.
The rejection-sampling selection in latent space resembles a mini-max choice; an interesting extension would be to make the selection differentiable or to use a learned similarity metric rather than L2 in latent space, potentially improving the stability-precedence tradeoff.

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

RDPO has a genuinely clever core idea: partially noise a real video, reverse-sample it with the frozen generator, and label the result 'preferred' over the model's own full-noise sample. That gives you preference pairs for DPO at zero annotation cost. The combination—reverse-sampling real latents plus selection for closeness to the model manifold—is new as far as I know, and it's an appealing way to inject physical priors.

The paper does a few things well. The human validation in Appendix A.1 confirms that the reverse-sampled clips are indeed judged more physically consistent than the model's own samples, which is the load-bearing assumption. The authors also test on two model sizes (LTX-2B open, Shopee-MUG-V-10B proprietary), and they are upfront about limitations. The self-citation to IPO is not a problem by itself; the overlapping authorship is fine when the cited result is real.

But the evidence doesn't yet establish the central claim that this is preference learning rather than distillation. Algorithm 1 never passes the text prompt to the denoiser—πθ is called with only a latent and a timestep—while Eq. (3) optimizes log πθ(x|c). If the preferred sample is generated without conditioning on c, the DPO loss is mis-specified. Maybe the implementation does condition on the prompt, but the paper needs to say so and ablate it. This is the first load-bearing soft spot.

Second, the isolation experiment in Table 1 is too weak to rule out the SFT/distillation reading. RDPO (w/o sft) is 24.71 PhysicsIQ versus 24.63 for SFT on real videos—a 0.08 gap with no error bars. Neither baseline trains on the partial-denoised positives that RDPO actually uses, so you cannot distinguish 'preference signal' from 'mild supervised regression to real videos.' Add a baseline that SFTs on the same x_preferred positives.

The other issues are standard but still need fixing: no significance tests or confidence intervals; the human-DPO comparison uses 1,000 labeled pairs versus 8,000 RDPO pairs, so it's an unfair data-scaling comparison; and the reverse-step schedule (42, 40, mixture) looks tuned on the very benchmarks reported. The 10B model results rely on a human study with no detail on rater count or agreement.

Bottom line: this is a promising direction and the core idea deserves a serious referee. But as written, the claims overreach. I'd send it to review and then demand major revision: clarify the prompt path in Algorithm 1, add a prompt-conditional ablation and an SFT-on-positives baseline, report error bars, and equalize data scales. If those checks hold up, RDPO could be a useful tool for physics-aware video alignment.

Referee Report

4 major / 5 minor

Summary. The paper proposes RDPO, an annotation-free post-training method for text-to-video diffusion models that aims to improve physical consistency. Preference pairs are constructed automatically by (i) generating a negative video from full Gaussian noise and (ii) generating a positive video by partially noising a real video and reverse-sampling from an intermediate timestep, with rejection sampling selecting the latent closest to the model's own sample. The pairs are used with a DPO-style loss in Eq. (3), interleaved with SFT, in a progressive curriculum that reduces the reverse-sampling step over iterations. Experiments on LTX-2B report improved PhysicsIQ and VBench scores, and experiments on Shopee-MUG-V-10B report favorable human evaluation and automatic metrics.

Significance. If the claims hold, RDPO is a practically valuable way to use real videos as preference supervision without human annotations or reward models, and the progressive curriculum is a sensible coarse-to-fine strategy. The paper has clear strengths: it provides an explicit algorithm, releases code, evaluates on two model scales, validates pair quality with a human study, and compares against human-annotated DPO. The main risk is whether the signal is genuinely conditional preference learning from real-video priors rather than a form of supervised reconstruction distillation; the current experiments do not fully separate these explanations.

major comments (4)

[Algorithm 1, Eq. (3)] The preference-pair construction pipeline never passes the text condition c through the sampling calls: lines 2, 3, 6, and 11 invoke f and πθ only with latents and timesteps, while the DPO objective in Eq. (3) evaluates log πθ(x_preferred|c). As written, x_preferred is not established to be a sample from the conditional policy πθ(·|c), so it is unclear what likelihood ratio the loss optimizes. Please specify how prompts condition reverse sampling, and include an ablation comparing unconditional versus prompt-conditional reverse sampling to verify transfer to text-conditioned generation.
[Section 4.1.2, Table 1] The 0.08-point PhysicsIQ gap between RDPO (w/o sft) and SFT (24.71 versus 24.63) is too small, and no error bars or significance tests are reported, to support the claim that preference learning from real-video latents adds something beyond supervised fitting. The comparison is also confounded: SFT is trained only on clean real videos, while RDPO's positives are partially denoised latents. Add an SFT baseline trained on the same x_preferred positives with the same data budget, and report variance across seeds or evaluation sets.
[Section 4.1.3, Table 2] The claim that annotation-free RDPO outperforms human-annotated DPO is not controlled for data quantity: RDPO uses 8,000 pairs versus 1,000 human-labeled pairs, and no confidence intervals are given. Please add a matched-size human DPO training run (or a learning-curve comparison) and report significance tests, since the observed differences on PhysicsIQ (25.21 versus 24.70) and Total Score (85.63 versus 85.44) are small.
[Section 4.1.4, Figure 3 and Table 3] The progressive schedule (42, 40, and a mixture of both) is chosen by inspecting Total Score on the same benchmarks used for final reporting, so the reported improvements are selected, not predicted, results. Please use a held-out validation set for schedule selection, or report all evaluated configurations and mark the selected one, to avoid selection bias in the headline numbers.

minor comments (5)

[Appendix A] The heading 'A Dissusion' contains a typo; it should read 'A Discussion'.
[Section A.1, Figure 6] The human validation study does not report the number of annotators, the number of pairs evaluated, or error bars on the accuracy curve; please add these details.
[Algorithm 2 and Table 1] Algorithm 2 unconditionally interleaves DPO and SFT every other step, but Table 1 reports a configuration called 'RDPO (w/o sft)'; please clarify how that configuration was trained.
[Section 4.2.2, Table 4] The comparison in Table 4 only contrasts Shopee-MUG-V-RDPO against the base Shopee-MUG-V; please clarify whether the column heading 'other open-source model' is intended to indicate additional baselines and, if so, why they are not reported.
[References] References [42] and [43] are the same paper; please remove the duplicate. Also, please provide a URL or release status for the WISA dataset if it is intended to be public.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the preference signal is external real-video data, the training objective is not the evaluation metric, and self-citations are not load-bearing.

full rationale

The derivation chain is not circular. RDPO's preference signal is an external ground truth: real videos. Algorithm 1 constructs x_preferred by partially noising and then denoising a real video latent, while x_model is produced by the same pretrained generator; the preference pair is therefore not derived from the target score (PhysicsIQ) or from the final model's outputs. Equation (3) is the standard DPO loss applied to these externally constructed pairs, and evaluation is done on held-out external benchmarks (VBench, PhysicsIQ) and human studies. No equation sets the training objective equal to the evaluation metric, and no parameter is fitted to the benchmark. The only author-overlapping citation is IPO (refs. [42]/[43] by Yang, Tan, and Li, with Tan and Li also on this paper), but it appears in a general enumeration of prior preference-optimization work and is not used to justify RDPO's core construction. The more substantive scientific concerns--that Algorithm 1's call signatures omit the prompt c that appears in Eq. (3), and that x_preferred is a near-reconstruction of the real video so the optimization signal may be closer to supervised distillation than to preference learning--are correctness and interpretation risks, not circularity, because the preference source and the evaluation remain external to the claimed result. Under the requirement to exhibit a specific reduction of a claimed result to its own inputs, no such reduction is present in the paper.

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

The method relies on real-video priors as the preference oracle; no new theoretical entities are introduced. All free parameters are standard training choices or step schedules.

free parameters (4)

Reverse sampling step t per training stage = 42, 40, mixed 42/40
Controls how much real-video information is injected into preferred samples; selected via ablations on the evaluation benchmarks (Fig 3).
Rejection sampling count K = not reported
Number of candidate real latents sampled per pair; affects the quality/selection of preferred latents but is omitted.
DPO temperature beta = not reported
Scales the preference loss; standard DPO hyperparameter, not specified.
LoRA rank / LR / batch size = 256, 1e-5, 256
Standard fine-tuning hyperparameters; not central to the method but required for reproduction.

assumptions (5)

domain assumption Reverse-sampled real video latents carry transferable physical priors
The method's positive samples are derived from partially noised real videos; the paper assumes denoising these yields physically better outputs (Section 3.2.1).
domain assumption L2 distance in latent space is a valid similarity metric for preference selection
Used for rejection sampling; no analysis of alternative metrics (Algorithm 1, line 7).
domain assumption Preference pairs are statistically distinguishable and align with human judgement
Supported by a human study (Figure 6) but only for a subset of conditions.
domain assumption The preferred and model samples share the same text condition c
The DPO objective (Eq. 3) conditions on c, but Algorithm 1 does not specify how prompts are applied to the reverse-sampled real videos.
domain assumption WISA and DPI datasets are representative of the physical laws of interest
The method relies on real data quality; the authors acknowledge this in Limitations (Section B).

how reviews work

0 comments

Cite this review

Pith. "Pith review of RDPO: Real Data Preference Optimization for Physics Consistency Video Generation." pith.science (2026). https://pith.science/paper/O4PYEWWY

@misc{pith2026250618655,
  author       = {Pith},
  title        = {Pith review of: RDPO: Real Data Preference Optimization for Physics Consistency Video Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/O4PYEWWY}},
  note         = {Machine review of arXiv:2506.18655}
}

read the original abstract

Video generation techniques have achieved remarkable advancements in visual quality, yet faithfully reproducing real-world physics remains elusive. Preference-based model post-training may improve physical consistency, but requires costly human-annotated datasets or reward models that are not yet feasible. To address these challenges, we present Real Data Preference Optimisation (RDPO), an annotation-free framework that distills physical priors directly from real-world videos. Specifically, the proposed RDPO reverse-samples real video sequences with a pre-trained generator to automatically build preference pairs that are statistically distinguishable in terms of physical correctness. A multi-stage iterative training schedule then guides the generator to obey physical laws increasingly well. Benefiting from the dynamic information explored from real videos, our proposed RDPO significantly improves the action coherence and physical realism of the generated videos. Evaluations on multiple benchmarks and human evaluations have demonstrated that RDPO achieves improvements across multiple dimensions. The source code and demonstration of this paper are available at: https://wwenxu.github.io/RDPO/

Figures

Figures reproduced from arXiv: 2506.18655 by the authors.

**Figure 2.** Visual comparison of video generation. Top: Videos generated by denoising from pure noise. Bottom: Videos generated by reverse sampling from an intermediate latent. Reverse sampling produces substantially better physical consistency and overall fidelity. 3.2.1 Annotation-Free Pairwise Data Generation Gradually decreasing the starting step of reverse sampling controls the noise level added to the real video, giving a… view at source ↗

**Figure 3.** Total score across different reverse steps [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

Figures from the paper (7 more)

**Figure 4.** Figure 4: Human evaluation of RDPO and Shopee-MUG-V model. [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]

**Figure 5.** Figure 5: Qualitative comparison between the Shopee-MUG-V model and our RDPO model. As [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]

**Figure 6.** Figure 6: Accuracy of RDPO-Constructed Preference Pairs Under Human Evaluation. The graph demonstrates increasing alignment between RDPO’s automated pair labeling and human judgments of physical consistency as the "positive" samples are derived using fewer reverse sampling ste…

**Figure 7.** Figure 7: Qualitative comparisons showcasing the impact of RDPO on the Shopee-MUG-V. RDPO [PITH_FULL_IMAGE:figures/full_fig_p013_7.png]

**Figure 8.** Figure 8: Qualitative comparisons showcasing the impact of RDPO on the Shopee-MUG-V, focusing [PITH_FULL_IMAGE:figures/full_fig_p014_8.png]

**Figure 9.** Figure 9: Qualitative comparisons showcasing the impact of RDPO on the LTX model, highlighting [PITH_FULL_IMAGE:figures/full_fig_p015_9.png]

**Figure 10.** Figure 10: Qualitative comparisons showcasing the impact of RDPO on the LTX model, demonstrat [PITH_FULL_IMAGE:figures/full_fig_p016_10.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

When Physical Preferences Meet Semantic Constraints: Physical and Semantic Direct Preference Optimization for Text-to-Video Generation
cs.CV 2026-07 conditional novelty 4.0 of 10

A semantic-conflict reweighting and staged-training version of DPO improves physical plausibility in text-to-video generation while partly preserving prompt semantics.

Reference graph

Works this paper leans on

42 extracted references · 10 canonical work pages · cited by 1 Pith paper

[1]

All are worth words: A vit backbone for diffusion models

Fan Bao et al. “All are worth words: A vit backbone for diffusion models”. In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2023, pp. 22669–22679

work page 2023
[2]

Kevin Black et al.Training Diffusion Models with Reinforcement Learning. 2024. arXiv: 2305.13301 [cs.LG].URL:https://arxiv.org/abs/2305.13301

arXiv 2024
[3]

Andreas Blattmann et al.Align your Latents: High-Resolution Video Synthesis with Latent Diffusion Models. 2023. arXiv: 2304.08818 [cs.CV] .URL: https://arxiv.org/abs/ 2304.08818

arXiv 2023
[4]

Video generation models as world simulators. 2024

Tim Brooks et al. “Video generation models as world simulators. 2024”. In:URL https://openai. com/research/video-generation-models-as-world-simulators3 (2024), p. 1

work page 2024
[5]

Genie: Generative interactive environments

Jake Bruce et al. “Genie: Generative interactive environments”. In:Forty-first International Conference on Machine Learning. 2024

work page 2024
[6]

Improving dynamic object interactions in text-to-video generation with ai feedback

Hiroki Furuta et al. “Improving dynamic object interactions in text-to-video generation with ai feedback”. In:arXiv preprint arXiv:2412.02617(2024)

arXiv 2024
[7]

I2v-adapter: A general image-to-video adapter for diffusion models

Xun Guo et al. “I2v-adapter: A general image-to-video adapter for diffusion models”. In:ACM SIGGRAPH 2024 Conference Papers. 2024, pp. 1–12

work page 2024
[9]

Animatediff: Animate your personalized text-to-image diffusion models without specific tuning

Yuwei Guo et al. “Animatediff: Animate your personalized text-to-image diffusion models without specific tuning”. In:arXiv preprint arXiv:2307.04725(2023)

arXiv 2023

Show all 42 references

[10]

Ltx-video: Realtime video latent diffusion

Yoav HaCohen et al. “Ltx-video: Realtime video latent diffusion”. In:arXiv preprint arXiv:2501.00103(2024)

2024 arXiv
[11]

Denoising diffusion probabilistic models

Jonathan Ho, Ajay Jain, and Pieter Abbeel. “Denoising diffusion probabilistic models”. In: Advances in neural information processing systems33 (2020), pp. 6840–6851

2020
[12]

Imagen video: high definition video generation with diffusion models (2022)

Jonathan Ho et al. “Imagen video: high definition video generation with diffusion models (2022)”. In:arXiv preprint arXiv:2210.02303(2022)

2022 arXiv
[13]

Cogvideo: Large-scale pretraining for text-to-video generation via trans- formers

Wenyi Hong et al. “Cogvideo: Large-scale pretraining for text-to-video generation via trans- formers”. In:arXiv preprint arXiv:2205.15868(2022)

2022 arXiv
[14]

Gen-Drive: Enhancing Diffusion Generative Driving Policies with Reward Modeling and Reinforcement Learning Fine-tuning

Zhiyu Huang et al. “Gen-Drive: Enhancing Diffusion Generative Driving Policies with Reward Modeling and Reinforcement Learning Fine-tuning”. In:arXiv preprint arXiv:2410.05582 (2024)

2024 arXiv
[15]

Vbench++: Comprehensive and versatile benchmark suite for video genera- tive models

Ziqi Huang et al. “Vbench++: Comprehensive and versatile benchmark suite for video genera- tive models”. In:arXiv preprint arXiv:2411.13503(2024)

2024 arXiv
[16]

How far is video generation from world model: A physical law perspective

Bingyi Kang et al. “How far is video generation from world model: A physical law perspective”. In:arXiv preprint arXiv:2411.02385(2024)

2024 arXiv
[17]

Hunyuanvideo: A systematic framework for large video generative models

Weijie Kong et al. “Hunyuanvideo: A systematic framework for large video generative models”. In:arXiv preprint arXiv:2412.03603(2024)

2024 arXiv
[18]

Kimin Lee et al.Aligning Text-to-Image Models using Human Feedback. 2023. arXiv: 2302. 12192 [cs.LG].URL:https://arxiv.org/abs/2302.12192

2023 arXiv
[19]

Aligning diffusion models by optimizing human utility

Shufan Li et al. “Aligning diffusion models by optimizing human utility”. In:arXiv preprint arXiv:2404.04465(2024)

2024 arXiv
[20]

Exploring the evolution of physics cognition in video generation: A survey

Minghui Lin et al. “Exploring the evolution of physics cognition in video generation: A survey”. In:arXiv preprint arXiv:2503.21765(2025)

2025 arXiv
[22]

Improving Video Generation with Human Feedback

Jie Liu et al. “Improving Video Generation with Human Feedback”. In:arXiv preprint arXiv:2501.13918(2025)

2025 arXiv
[23]

Flow straight and fast: Learning to generate and transfer data with rectified flow

Xingchao Liu, Chengyue Gong, and Qiang Liu. “Flow straight and fast: Learning to generate and transfer data with rectified flow”. In:arXiv preprint arXiv:2209.03003(2022)

2022 arXiv
[24]

Sdedit: Guided image synthesis and editing with stochastic differential equations

Chenlin Meng et al. “Sdedit: Guided image synthesis and editing with stochastic differential equations”. In:arXiv preprint arXiv:2108.01073(2021)

2021 arXiv
[25]

Do generative video models learn physical principles from watching videos?

Saman Motamed et al. “Do generative video models learn physical principles from watching videos?” In:arXiv preprint arXiv:2501.09038(2025). 17

2025 arXiv
[26]

Scalable diffusion models with transformers

William Peebles and Saining Xie. “Scalable diffusion models with transformers”. In:Pro- ceedings of the IEEE/CVF international conference on computer vision. 2023, pp. 4195– 4205

2023
[27]

Video diffusion alignment via reward gradients

Mihir Prabhudesai et al. “Video diffusion alignment via reward gradients”. In:arXiv preprint arXiv:2407.08737(2024)

2024 arXiv
[28]

High-resolution image synthesis with latent diffusion models

Robin Rombach et al. “High-resolution image synthesis with latent diffusion models”. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2022, pp. 10684–10695

2022
[29]

Make-a-video: Text-to-video generation without text-video data

Uriel Singer et al. “Make-a-video: Text-to-video generation without text-video data”. In:arXiv preprint arXiv:2209.14792(2022)

2022 arXiv
[30]

Deep Unsupervised Learning using Nonequilibrium Thermody- namics

Jascha Sohl-Dickstein et al. “Deep Unsupervised Learning using Nonequilibrium Thermody- namics”. en-US. In:arXiv: Learning,arXiv: Learning(Mar. 2015)

2015
[31]

Denoising diffusion implicit models

Jiaming Song, Chenlin Meng, and Stefano Ermon. “Denoising diffusion implicit models”. In: arXiv preprint arXiv:2010.02502(2020)

2020 arXiv
[32]

Diffusion model alignment using direct preference optimization

Bram Wallace et al. “Diffusion model alignment using direct preference optimization”. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2024, pp. 8228–8238

2024
[33]

WISA: World simulator assistant for physics-aware text-to-video generation

Jing Wang et al. “WISA: World simulator assistant for physics-aware text-to-video generation”. In:arXiv preprint arXiv:2503.08153(2025)

2025 arXiv
[34]

Modelscope text-to-video technical report

Jiuniu Wang et al. “Modelscope text-to-video technical report”. In:arXiv preprint arXiv:2308.06571(2023)

2023 arXiv
[35]

Fairy: Fast parallelized instruction-guided video-to-video synthesis

Bichen Wu et al. “Fairy: Fast parallelized instruction-guided video-to-video synthesis”. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2024, pp. 8261–8270

2024
[36]

Godiva: Generating open-domain videos from natural descriptions

Chenfei Wu et al. “Godiva: Generating open-domain videos from natural descriptions”. In: arXiv preprint arXiv:2104.14806(2021)

2021 arXiv
[37]

Boosting text-to-video generative model with MLLMs feedback

Xun Wu et al. “Boosting text-to-video generative model with MLLMs feedback”. In:The Thirty-eighth Annual Conference on Neural Information Processing Systems. 2024

2024
[38]

A survey on video diffusion models

Zhen Xing et al. “A survey on video diffusion models”. In:ACM Computing Surveys57.2 (2024), pp. 1–42

2024
[39]

Jiazheng Xu et al.VisionReward: Fine-Grained Multi-Dimensional Human Preference Learn- ing for Image and Video Generation. 2025. arXiv: 2412 . 21059 [cs.CV].URL: https : //arxiv.org/abs/2412.21059

2025 arXiv
[40]

Phyt2v: Llm-guided iterative self-refinement for physics-grounded text-to- video generation

Qiyao Xue et al. “Phyt2v: Llm-guided iterative self-refinement for physics-grounded text-to- video generation”. In:arXiv preprint arXiv:2412.00596(2024)

2024 arXiv
[41]

Learning interactive real-world simulators

Mengjiao Yang et al. “Learning interactive real-world simulators”. In:arXiv preprint arXiv:2310.061141.2 (2023), p. 6

2023 arXiv
[43]

Ipo: Iterative preference optimization for text-to- video generation

Xiaomeng Yang, Zhiyu Tan, and Hao Li. “Ipo: Iterative preference optimization for text-to- video generation”. In:arXiv preprint arXiv:2502.02088(2025)

2025
[44]

InstructVideo: Instructing Video Diffusion Models with Human Feedback

Hangjie Yuan et al. “InstructVideo: Instructing Video Diffusion Models with Human Feedback”. In:Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). June 2024, pp. 6463–6474

2024
[45]

gddim: Generalized denoising diffusion implicit models

Qinsheng Zhang, Molei Tao, and Yongxin Chen. “gddim: Generalized denoising diffusion implicit models”. In:arXiv preprint arXiv:2206.05564(2022). 18

2022 arXiv

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [1]

All are worth words: A vit backbone for diffusion models

Fan Bao et al. “All are worth words: A vit backbone for diffusion models”. In:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2023, pp. 22669–22679

work page 2023

[2] [2]

Kevin Black et al.Training Diffusion Models with Reinforcement Learning. 2024. arXiv: 2305.13301 [cs.LG].URL:https://arxiv.org/abs/2305.13301

arXiv 2024

[3] [3]

Andreas Blattmann et al.Align your Latents: High-Resolution Video Synthesis with Latent Diffusion Models. 2023. arXiv: 2304.08818 [cs.CV] .URL: https://arxiv.org/abs/ 2304.08818

arXiv 2023

[4] [4]

Video generation models as world simulators. 2024

Tim Brooks et al. “Video generation models as world simulators. 2024”. In:URL https://openai. com/research/video-generation-models-as-world-simulators3 (2024), p. 1

work page 2024

[5] [5]

Genie: Generative interactive environments

Jake Bruce et al. “Genie: Generative interactive environments”. In:Forty-first International Conference on Machine Learning. 2024

work page 2024

[6] [6]

Improving dynamic object interactions in text-to-video generation with ai feedback

Hiroki Furuta et al. “Improving dynamic object interactions in text-to-video generation with ai feedback”. In:arXiv preprint arXiv:2412.02617(2024)

arXiv 2024

[7] [7]

I2v-adapter: A general image-to-video adapter for diffusion models

Xun Guo et al. “I2v-adapter: A general image-to-video adapter for diffusion models”. In:ACM SIGGRAPH 2024 Conference Papers. 2024, pp. 1–12

work page 2024

[8] [9]

Animatediff: Animate your personalized text-to-image diffusion models without specific tuning

Yuwei Guo et al. “Animatediff: Animate your personalized text-to-image diffusion models without specific tuning”. In:arXiv preprint arXiv:2307.04725(2023)

arXiv 2023