Ouroboros-Diffusion: Exploring Consistent Content Generation in Tuning-free Long Video Diffusion
Open accessAbstract
The first-in-first-out (FIFO) video diffusion, built on a pre-trained text-to-video model, has recently emerged as an effective approach for tuning-free long video generation. This technique maintains a queue of video frames with progressively increasing noise, continuously producing clean frames at the queue's head while Gaussian noise is enqueued at the tail. However, FIFO-Diffusion often struggles to keep long-range temporal consistency in the generated videos due to the lack of correspondence modeling across frames. In this paper, we propose Ouroboros-Diffusion, a novel video denoising framework designed to enhance structural and content (subject) consistency, enabling the generation of consistent videos of arbitrary length. Specifically, we introduce a new latent sampling technique at the queue tail to improve structural consistency, ensuring perceptually smooth transitions among frames. To enhance subject consistency, we devise a Subject-Aware Cross-Frame Attention (SACFA) mechanism, which aligns subjects across frames within short segments to achieve better visual coherence. Furthermore, we introduce self-recurrent guidance. This technique leverages information from all previous cleaner frames at the front of the queue to guide the denoising of noisier frames at the end, fostering rich and contextual global information interaction. Extensive experiments of long video generation on the VBench benchmark demonstrate the superiority of our Ouroboros-Diffusion, particularly…
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
8 of 12 cited sources checked · 4 supported · 2 flagged · accuracy not yet measured
- Premisecitation
FIFO-Diffusion often struggles to keep long-range temporal consistency in generated videos due to the lack of correspondence modeling across frames.
“However, FIFO-Diffusion often struggles to keep longrange temporal consistency in the generated videos due to the lack of correspondence modeling across frames.”
Quote in the paperSubstantial claim1 of 1 sources checked- cites Kim et al. 2024 source
Section 3.1 (p. 4) says FIFO-Diffusion propagates context to later frames and lets each frame reference preceding frames, extending consistency to longer sequences. The qualitative results (p. 6) report that it maintains motion consistency by referencing earlier frames. The paper attributes difficulty maintaining long-term context to chunked autoregressive methods, not FIFO-Diffusion; it does not identify a lack of cross-frame correspondence modeling as the cause. (LLM, conf=0.88)
Checked by a reasoning model — may be wrong.
Experimental — accuracy not yet measured
- Premisecitation
The tail latent is typically sampled from a standard Gaussian distribution, which contains pure noise without any visual information.
“The tail latent is typically sampled from a standard Gaussian distribution, which contains pure noise without any visual information.”
Quote in the paperSubstantial claim1 of 1 sources checked- cites Kim et al. 2024 source
- Premisecitation
The low-frequency component preserves layout information for overall video consistency.
“The low-frequency component preserves layout information for overall video consistency, while the high-frequency component introduces necessary video dynamics.”
Quote in the paperSubstantial claim1 of 2 sources checked- cites Wu et al. 2023b
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- cites Everaert et al. 2024 source
Section 3.2 says low-frequency information from an image can survive diffusion noise, and Section 5.4 shows that selecting low-frequency components can influence generated images. The paper discusses image brightness and color, however, not preservation of layout across video frames or overall video consistency. (LLM, conf=0.82)
Experimental — accuracy not yet measured
- Premisecitation
The low-frequency component in latent space primarily corresponds to layout and overall structure in pixel space.
“Recent advances (Wu et al. 2023b;Everaert et al. 2024) indicate that the lowfrequency component in the latent space primarily corresponds to the layout and overall structure in the pixel space.”
Quote in the paperSubstantial claim1 of 2 sources checked- cites Wu et al. 2023b
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- cites Everaert et al. 2024 source
Sections 4.2.2 and 5.3 link low-frequency components of Stable Diffusion latents to generated-image brightness and colour variation across image areas. Section 5.4 shows that selecting those components can bias brightness and colour, but our copy does not establish that they primarily correspond to layout or overall structure in pixel space. The passages do not give page numbers. (LLM, conf=0.83)
Experimental — accuracy not yet measured
- Findingown data
Directly applying noise to the second-to-last frame latent and using it as the new tail latent results in generated videos with limited dynamics due to excessive similarities in visual content.
“However, we discover that this approach results in generated videos with limited dynamics due to the excessive similarities in visual content.”
Quote in the paperSubstantial claim
- Findingown data
Extensive experiments of long video generation on VBench demonstrate the superiority of Ouroboros-Diffusion, particularly in subject consistency, motion smoothness, and temporal consistency.
“experiments of long video generation on the VBench benchmark demonstrate the superiority of our Ouroboros-Diffusion, particularly in terms of subject consistency, motion smoothness, and temporal consistency.”
- Also stated in: § 1 Introduction · plain
“Extensive experiments on VBench verify the effectiveness of our proposal in terms of both visual and motion quality.”
Quote in the paperSubstantial claim - Also stated in: § 1 Introduction · plain
- Findingreasoning
The appearance of the cat changes significantly in the output video of Figure 1.
“For instance, the appearance of the cat changes significantly in the output video of Figure 1.”
Quote in the paperSubstantial claim - Findingown data
Using the head frame's low-frequency component for latent sampling improves Motion Smoothness from 95.88% to 97.51%.
“Using the head frame's lowfrequency component for latent sampling improves it from 95.88% to 97.51%.”
Quote in the paperSubstantial claim - Findingown data
Integrating coherent tail latent sampling into the base model (A) yields a significant performance boost (1.52%) in Subject Consistency for model B.
“When integrating coherent tail latent sampling into the base model (A), a significant performance boost (1.52%) is attained by model B in Subject Consistency.”
Quote in the paperNot a substantial claim - Findingown data
Ouroboros-Diffusion consistently outperforms other baselines across various metrics on VBench for single-scene long video generation.
“Overall, Ouroboros-Diffusion consistently outperforms other baselines across various metrics.”
Quote in the paperSubstantial claim - Findingown data
The pre-trained base model VideoCrafter2 achieves a dynamic degree score of 42.01.
“To validate this, we calculate the dynamic degree of the pre-trained base model VideoCrafter2 (Chen et al. 2024a), which achieves a score of 42.01.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites Chen et al. 2024a
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Rests on an assumption: The dynamic degree metric validly measures motion variability.
- Findingown data
Independent Gaussian noise yields the lowest Motion Smoothness.
“As expected, independent Gaussian noise yields the lowest Motion Smoothness.”
Quote in the paperNot a substantial claim - Findingown data
Whether using historic frame guidance (λ=1) or current frame guidance (λ=0) at the queue head, there are notable performance improvements in Subject Consistency and Motion Smoothness.
“As shown in Table 5, whether using historic frame guidance (λ=1) or the current frame guidance (λ=0) at the queue head, there are notable performance improvements in Subject Consistency and Motion Smoothness.”
Quote in the paperSubstantial claim - Findingown data
By adjusting enqueued latents with tail information, Ouroboros-Diffusion further enhances motion quality.
“By adjusting enqueued latents with tail information, Ouroboros-Diffusion further enhances motion quality.”
Quote in the paperSubstantial claim - Findingown data
Ouroboros-Diffusion attains a dynamic degree of 44.12.
“Ouroboros-Diffusion attains a higher dynamic degree of 44.12, confirming that our method maintains motion variability while further enhancing content consistency in video synthesis.”
Quote in the paperSubstantial claim- Rests on an assumption: The dynamic degree metric validly measures motion variability.
- Findingown data
Model D (Ouroboros-Diffusion) achieves the best performance among the ablated models.
“Finally, model D, (i.e., our Ouroboros-Diffusion), achieves the best performance”
Quote in the paperSubstantial claim - Findingown data
By enhancing subject token alignment through subject-aware cross-frame attention, model C outperforms model B across all metrics.
“By enhancing subject token alignment through subject-aware cross-frame attention, model C outperforms model B across all metrics.”
Quote in the paperSubstantial claim - Premisenone given
In diagonal denoising, global visual consistency is not explicitly considered during the diffusion process.
“In the design of diagonal denoising, the global visual consistency is not explicitly considered during the diffusion process.”
Quote in the paperSubstantial claim - Interpretationreasoning
Coherent tail latent sampling allows for consistent yet dynamic visual continuation, ensuring a smooth transition from Qτ to Qτ+1.
“The coherent tail latent allows for consistent yet dynamic visual continuation, ensuring a smooth transition from Q τ to Q τ +1 .”
Quote in the paperSubstantial claim - Interpretationnone given
Self-recurrent guidance leverages information from all previous cleaner frames at the front of the queue to guide the denoising of noisier frames at the end, fostering rich and contextual global information interaction.
“Furthermore, we introduce self-recurrent guidance.This technique leverages information from all previous cleaner frames at the front of the queue to guide the denoising of noisier frames at the end, fostering rich and contextual global information interaction.”
Quote in the paperNot a substantial claim- Rests on an assumption: The first f cleaner frame latents of the video sequence contain clearer and more critical visual information, making them an essential basis for the construction of long-term memory.
- Interpretationreasoning
Coherent tail latent sampling improves structural consistency, ensuring perceptually smooth transitions among frames.
“Specifically, we introduce a new latent sampling technique at the queue tail to improve structural consistency, ensuring perceptually smooth transitions among frames.”
Quote in the paperSubstantial claim- Rests on an assumption: The high-frequency component of random noise introduces necessary video dynamics.
- Premisenone given
Independently enqueued Gaussian noise can lead to content discrepancies between frame latents near the tail of the queue.
“However, the independently enqueued Gaussian noise can lead to content discrepancies between the frame latents near the tail of the queue.”
Quote in the paperNot a substantial claim - Interpretationreasoning
SACFA aligns subjects across frames within short segments to achieve better visual coherence.
“To enhance subject consistency, we devise a Subject-Aware Cross-Frame Attention (SACFA) mechanism, which aligns subjects across frames within short segments to achieve better visual coherence.”
Quote in the paperSubstantial claim - Interpretationreasoning
SACFA strengthens local subject correspondence by allowing subject information from neighboring frames to influence current frames, ensuring consistent subject representation.
“SACFA strengthens local subject correspondence by allowing subject information from neighboring frames to influence current frames, ensuring consistent subject representation.”
Quote in the paperSubstantial claim - Interpretationreasoning
Self-recurrent guidance aligns generated latents more closely with the subject features of previously generated frames, thereby enhancing long-range subject consistency in the synthesized video.
“By integrating this term, the self-recurrent guidance aligns the generated latents more closely with the subject features of the previously generated frames, thereby enhancing longrange subject consistency in the synthesized video.”
Quote in the paperSubstantial claim- Rests on an assumption: The first f cleaner frame latents of the video sequence contain clearer and more critical visual information, making them an essential basis for the construction of long-term memory.
- Interpretationnone given
Ouroboros-Diffusion is a tuning-free long video generation framework.
“we introduce a novel denoising framework termed Ouroboros-Diffusion, designed for tuning-free long video generation.”
Quote in the paperSubstantial claim - Interpretationown data
The highest frame consistency (Temporal Flickering) demonstrates the effectiveness of coherent tail latent sampling, which enforces similarity in image layout between adjacent frames to enhance structural consistency.
“The highest frame consistency, as indicated by the Temporal Flickering metric, demonstrates the effectiveness of our coherent tail latent sampling, which enforces similarity in image layout between adjacent frames to enhance structural consistency.”
Quote in the paperSubstantial claim - Interpretationown data
These results highlight the advantage of guiding the denoising process using subject information through gradient-based adjustments.
“These results highlight the advantage of guiding the denoising process using subject information through gradient-based adjustments.”
Quote in the paperSubstantial claim - Interpretationreasoning
A discrepancy between the visual information in the tail latent and its neighboring latents leads to inconsistencies and frame flickering during video generation.
“This creates a discrepancy between the visual information in the tail latent and its neighboring latents, leading to inconsistencies.As a result, the model may struggle to reconcile these differences, often resulting in frame flickering during video generation.”
Quote in the paperSubstantial claim - Interpretationown data
The best performance is achieved by recurrently propagating subject information from the queue head to the tail frames for latent optimization during video denoising.
“by recurrently propagating the subject information of frames from the queue head to the tail frames for latent optimization during video denoising.”
Quote in the paperSubstantial claim - Interpretationreasoning
The appearance gap between queue head and tail latents limits structural guidance.
“However, the appearance gap between queue head and tail latents limits structural guidance.”
Quote in the paperSubstantial claim - Interpretationreasoning
The base model (FIFO-Diffusion) has a weakness where structural information may be overlooked due to the enqueueing of independent Gaussian noise at the queue tail.
“This highlights a weakness of the base model (i.e., FIFO-Diffusion), where structural information may be overlooked due to the enqueueing of independent Gaussian noise at the queue tail.”
Quote in the paperSubstantial claim - Premisenone given
Neighboring latents have already undergone partial denoising in previous steps, introducing some degree of visual content.
“Meanwhile, neighboring latents have already undergone partial denoising in previous steps, introducing some degree of visual content.”
Quote in the paperSubstantial claim - Interpretationown data
Using information guidance within the queue enhances visual alignment in diffusion.
“demonstrating the advantage of using information guidance within the queue to enhance visual alignment in diffusion.”
Quote in the paperSubstantial claim
- May combine two claims. 16
- Classified as probably not a substantial claim. 4
- The quote only partly states this claim. 2
May combine two claims.
PremisecitationFIFO-Diffusion fails to utilize visual information contained in earlier denoised frames, which exacerbates temporal inconsistency in continuous video diffusion.
“Moreover, FIFO-Diffusion fails to utilize visual information contained in the earlier denoised frames, which exacerbates temporal inconsistency in continuous video diffusion.”
Quote in the paperSubstantial claim1 of 1 sources checked- cites Kim et al. 2024 source
Section 3.1 (p. 4) says the opposite: FIFO-Diffusion propagates context to later frames, lets each frame reference preceding frames, and thereby extends local consistency to longer sequences. Section 1 (p. 2) also says its lookahead denoising lets frames reference cleaner frames. The paper attributes inadequate contextual information and resulting discontinuities to chunked autoregressive methods, not FIFO-Diffusion. (LLM, conf=0.89)
Checked by a reasoning model — may be wrong.
Experimental — accuracy not yet measured
Probably not a substantial claim: may not pass S3 (load-bearing).
PremisecitationThe diagonal denoising approach proposed by Kim et al. (2024) maintains a queue of frames with progressively increasing noise levels, enabling frame-by-frame generation at each step.
“Compared to general video diffusion that all frames share the same noise level, the diagonal denoising approach proposed by Kim et al. (2024) maintains a queue of frames with progressively increasing noise levels, enabling frame-by-frame generation at each step.”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites Kim et al. 2024 source
Classified as probably not a substantial claim.
PremisecitationThe latent enqueue-dequeue cycle allows for the incremental generation of video frames.
“The latent enqueue-dequeue cycle allows for the incremental generation of video frames.”
Quote in the paperSubstantial claim1 of 1 sources checked- cites Kim et al. 2024 source
May combine two claims.
PremisecitationOnce frames are fully denoised, they are dequeued and no longer contribute to the generation of subsequent frames.
“Once frames are fully denoised, they are dequeued and no longer contribute to the generation of subsequent frames.”
Quote in the paperSubstantial claim1 of 1 sources checked- cites Kim et al. 2024 source
May combine two claims.
Findingown dataOuroboros-Diffusion successfully generates smoother transitions and more consistent visual content in multi-scene long video generation.
“As shown, Ouroboros-Diffusion successfully generates smoother transitions (e.g., scene changes with the astronaut maintaining the same motion direction) and more consistent visual content (e.g., a single, identical astronaut rather than two).”
Quote in the paperSubstantial claimMay combine two claims.
Findingown dataExperiments on the VBench benchmark validate the effectiveness of Ouroboros-Diffusion, demonstrating improvements in both visual quality and motion smoothness.
“Experiments on the VBench benchmark validate the effectiveness of Ouroboros-Diffusion, demonstrating improvements in both visual quality and motion smoothness.”
Quote in the paperSubstantial claimMay combine two claims.
Findingown dataFIFO-Diffusion maintains some content coherence but its enqueued independent Gaussian noise results in background variations.
“Although the FIFO-Diffusion denoising strategy maintains some content coherence (e.g., the appearance of the cat), the enqueued independent Gaussian noise still results in background variations (e.g., changes in the building behind the pool).”
Quote in the paperSubstantial claimMay combine two claims.
Findingown dataOuroboros-Diffusion outperforms all baselines in Subject/Background Consistency, Motion Smoothness, and Temporal Flickering on multi-scene VBench.
“Ouroboros-Diffusion outperforms all baselines in Subject/Background Consistency, Motion Smoothness, and Temporal Flickering.”
Quote in the paperSubstantial claimMay combine two claims.
Findingown dataStreamingT2V and FreeNoise often generate unreasonable or inconsistent content.
“For example, StreamingT2V and FreeNoise often generate unreasonable or inconsistent content (e.g., the change of red collar in FreeNoise).”
Quote in the paperSubstantial claimThe quote only partly states this claim.
Findingown dataOuroboros-Diffusion demonstrates substantial performance boosts (0.66%∼1.77%) in Subject Consistency on multi-scene VBench.
“Specifically, our approach demonstrates substantial performance boosts (0.66%∼1.77%) in Subject Consistency.”
Quote in the paperSubstantial claimMay combine two claims.
Findingown dataOuroboros-Diffusion achieves a Temporal Flickering score of 96.12% on single-scene VBench, surpassing FIFO-Diffusion by 2.74% and FreeNoise by 2.50%.
“Notably, Ouroboros-Diffusion achieves a Temporal Flickering score of 96.12%, surpassing the tuning-free approaches FIFO-Diffusion and FreeNoise by 2.74% and 2.50%, respectively.”
Quote in the paperSubstantial claimMay combine two claims.
Findingown dataCompared to other baselines, Ouroboros-Diffusion consistently produces videos with more seamless transitions and superior visual consistency.
“Compared to other baselines, Ouroboros-Diffusion consistently produces videos with more seamless transitions and superior visual consistency.”
Quote in the paperSubstantial claimMay combine two claims.
Findingown dataOuroboros-Diffusion preserves both subject and background consistency effectively.
“In contrast, the video generated by our Ouroboros-Diffusion preserves both subject and background consistency effectively”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S4 (based).
Findingnone givenOuroboros-Diffusion does not compromise video motion strength to improve temporal consistency.
“It is important to note that our approach does not compromise video motion strength (e.g., causing static video generation) to improve temporal consistency.”
Quote in the paperNot a substantial claim- Rests on an assumption: The dynamic degree metric validly measures motion variability.
May combine two claims.
InterpretationreasoningLater frames cannot reference information from previously generated frames, which causes the visual appearance of backgrounds and objects to gradually shift during continuous frame denoising, resulting in inconsistencies over time.
“Consequently, later frames cannot reference the information from previously generated frames, which causes the visual appearance of backgrounds and objects to gradually shift during continuous frame denoising, resulting in inconsistencies over time.”
Quote in the paperSubstantial claimMay combine two claims.
InterpretationreasoningOuroboros-Diffusion enhances structural and content (subject) consistency, enabling the generation of consistent videos of arbitrary length.
“In this paper, we propose Ouroboros-Diffusion, a novel video denoising framework designed to enhance structural and content (subject) consistency, enabling the generation of consistent videos of arbitrary length.”
Quote in the paperSubstantial claim- Rests on an assumption: The high-frequency component of random noise introduces necessary video dynamics.
Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.
PremisecitationFreeNoise exploits a noise scheduler for multi-scene video generation, but it emphasizes prompt adjustment in different denoising steps for motion injection.
“Note that FreeNoise exploits a noise scheduler for multi-scene video generation, but it emphasizes on prompt adjustment in different denoising steps for motion injection.”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites Qiu et al. 2024
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
May combine two claims.
Interpretationnone givenOuroboros-Diffusion utilizes long-term memory derived from the frame at the head of the queue to guide the denoising of noisier frames near the tail, optimizing the latent through a subject-aware gradient during video denoising.
“Specifically, Ouroboros-Diffusion utilizes the long-term memory derived from the frame at the head of the queue to guide the denoising of noisier frames near the tail, optimizing the latent through a subject-aware gradient during video denoising.”
Quote in the paperSubstantial claim- Rests on an assumption: The first f cleaner frame latents of the video sequence contain clearer and more critical visual information, making them an essential basis for the construction of long-term memory.
May combine two claims.
Interpretationown dataThe best performances of Subject Consistency (96.06%) and Background Consistency (96.90%) show that Ouroboros-Diffusion benefits from subject-level guidance, resulting in natural coherence throughout long video generation.
“Additionally, the best performances of Subject Consistency (96.06%) and Background Consistency (96.90%) further show that Ouroboros-Diffusion benefits from subject-level guidance, resulting in natural coherence throughout long video generation.”
Quote in the paperSubstantial claimMay combine two claims.
Interpretationown dataOuroboros-Diffusion maintains motion variability while further enhancing content consistency in video synthesis.
“confirming that our method maintains motion variability while further enhancing content consistency in video synthesis.”
Quote in the paperSubstantial claim- Rests on an assumption: The dynamic degree metric validly measures motion variability.
The quote only partly states this claim.
PremisereasoningThe subject consistency of videos produced by diagonal denoising is often compromised due to lack of global consistency modeling and latent discrepancy near the tail.
“However, the subject consistency of the videos produced by diagonal denoising is often compromised due to the following two limitations: Lack of Global Consistency Modeling.”
Quote in the paperSubstantial claim
The paper relies on this without support.
Most current video diffusion models are trained on short clips (e.g., 16 frames), which is a significant challenge when scaling to long video generation.
“However, most current video diffusion models are trained on short clips (e.g., 16 frames), a significant challenge when scaling to long video generation.”
The paper relies on this without support.
The first f cleaner frame latents of the video sequence contain clearer and more critical visual information, making them an essential basis for the construction of long-term memory.
“These initial f frames are particularly valuable as they contain clearer and more critical visual information, making them an essential basis for the construction of long-term memory.”
The paper relies on this without support.
The high-frequency component of random noise introduces necessary video dynamics.
“The low-frequency component preserves layout information for overall video consistency, while the high-frequency component introduces necessary video dynamics.”
The paper relies on this without support.
The dynamic degree metric validly measures motion variability.
“we calculate the dynamic degree of the pre-trained base model VideoCrafter2”
“Ouroboros-Diffusion attains a higher dynamic degree of 44.12”
The paper relies on this without support.
The low-frequency component of latents near the queue tail carries structural information that can enhance structural consistency.
“inject structural information into newly enqueued Gaussian noise by leveraging the low-frequency component of latents near the queue tail, thereby enhancing the consistency of the overall structural layout.”
Cites
The publisher hasn’t deposited this paper’s reference list.
Research works printed but not linked: 5
- [13]13. Ho, J.; Salimans, T.; Gritsenko, A.; Chan, W.; Norouzi, M.; and Fleet, D. J. 2022b. Video Diffusion Models. In NeurIPS. Hong, W.; Ding, M.; Zheng, W.; Liu, X.; and Tang, J. 2023. CogVideo: Large-Scale Pretraining for Text-to-Video Gen- eration via Transformers. In ICLR.
- [25]25. Meng, C.; He, Y.; Song, Y.; Song, J.; Wu, J.; Zhu, J.-Y.; and Ermon, S. 2022. SDEdit: Guided Image Synthesis and Edit- ing with Stochastic Differential Equations. In ICLR.
- [30]30. Radford, A.; Kim, J. W.; Hallacy, C.; Ramesh, A.; Goh, G.; Agarwal, S.; Sastry, G.; Askell, A.; Mishkin, P.; Clark, J.; et al. 2021. Learning Transferable Visual Models from Nat- ural Language Supervision. In ICML.
- [31]31. Singer, U.; Polyak, A.; Hayes, T.; Yin, X.; An, J.; Zhang, S.; Hu, Q.; Yang, H.; Ashual, O.; Gafni, O.; et al. 2023. Make-A-Video: Text-to-Video Generation without Text- Video Data. In ICLR.
- [38]38. Wang, X.; Yuan, H.; Zhang, S.; Chen, D.; Wang, J.; Zhang, Y.; Shen, Y.; Zhao, D.; and Zhou, J. 2023b. VideoComposer: Compositional Video Synthesis with Motion Controllability. In NeurIPS. Wu, J. Z.; Ge, Y.; Wang, X.; Lei, W.; Gu, Y.; Shi, Y.; Hsu, W.; Shan, Y.; Qie, X.; and Shou, M. Z. 2023a. Tune-A-Video: One-Shot Tuning of Image Diffusion Models for Text-to- Video Generation. In ICCV.
Cited by
Sorted by citations
Video diffusion generation: comprehensive review and open problems
Ma, W., Yang, X., Jiao, L., Li, L., Liu, X., Liu, F., Chen, P., Yang, Y., Ma, M., Sun, L., et al. (2025) · Artificial Intelligence Review · doi:10.1007/s10462-025-11331-6 · Cited by 12 · see citing works
Training-Free and Adaptive Sparse Attention for Efficient Long Video Generation
Xia, Y., Ling, S., Fu, F., Wang, Y., Li, H., Xiao, X., & Cui, B. (2025) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv51701.2025.01483 · Cited by 7 · see citing works
Aligning Global Semantics and Local Textures in Generative Video Enhancement*
Chen, Z., Long, F., Qiu, Z., Yao, T., Zhou, W., Luo, J., & Mei, T. (2025) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv51701.2025.01587 · Cited by 1 · see citing works
Self-citationOn-Policy Diffusion Reinforcement Learning Meets Off-Policy Quality Anchoring
Liu, Z., Qiu, Z., Xie, Y., Pan, Y., Yao, T., & Mei, T. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-37577-3_31
Self-citationFan, C.-D., Mu, C.-W. T., Chang, C. C., Lin, C.-Y., Wu, K.-R., Tseng, Y., & Liu, Y.-L. (2026) · ACM SIGGRAPH Conference (SIGGRAPH) · doi:10.1145/3799902.3811151
- Authors
- Chen, J., Long, F., An, J., Qiu, Z., Yao, T., Luo, J., & Mei, T.
- Affiliations
- University of Rochester (US)
- Venue
- Proceedings of the AAAI Conference on Artificial Intelligence
- Published
- 2025-04-11
- Volume
- 39
- Issue
- 2
- Pages
- 2079-2087
- ISSN
- 2159-5399 (ISSN-L)
- 2374-3468
- OpenAlex
- W4409346405
- Cited by
- 5
- References
- 0
- Citations by year
- 2026: 2 · 2025: 3
- FWCI
- 19.26
- Type
- conference paper
- Language
- en
- Open access
- diamond
- doi.org
- Locations
- Proceedings of the AAAI Conference on Artificial Intelligence (publishedVersion)
- AAAI Publications (The Association for the Advancement of Artificial Intelligence (AAAI)) (submittedVersion)
- Topics
- Social Sciences › Sociology and Political Science › Multimedia Communication and Technology
- Keywords
- long video generation, video denoising, temporal coherence, text-to-video models
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-11.
Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-11.
Cited sources are checked by gpt-6-sol.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
