&quot;Wednesday&quot; and &quot;Scream&quot; star Jenna Ortega’s latest movie — which was a flop in U.S. theaters — is making its streaming debut on Netflix. Find out when it will premiere on the streamer.

Jenna Ortega Movie Flop To Make Streaming Debut On Netflix

Netflix is home to hits like Damsel and Society of the Snow, but this new film is the streamer&#x27;s best 2024 movie so far. Here&#x27;s why you need to watch it now.

Netflix's most addictive movie of 2024 is now streaming. Here's why you need to watch it | Digital Trends

&quot;The people who hate it HATE IT. I loved it. It was the first time I actually felt like I understood everyone&#x27;s love of Star Wars.&quot;

People Are Sharing The Widely Hated Movies They Secretly Love

“I’ve left them very much to do their own thing”, she said

Emily Maitlis breaks silence on new Netflix drama Scoop based on her interview with Prince Andrew | The Independent

Immerse yourself in the shadowy world of crime and intrigue with our list of the best film noir movies. Explore the timeless classics that define this genre.

The 25 Best Film Noir Movies Of All Time

Netflix&#x27;s latest batch of new movies packs critically-acclaimed flicks including &quot;King Richard&quot;, &quot;Baby Driver&quot; and more.

5 new to Netflix movies with 90% or higher on Rotten Tomatoes | Tom's Guide

Disney has announced release dates for &#x27;The Mandalorian &amp; Grogu,&#x27; &#x27;Toy Story 5,&#x27; its live-action version of &#x27;Moana&#x27; and its &#x27;Tron&#x27; sequel.

Disney Sets 'Star Wars' Movie 'The Mandalorian & Grogu,' 'Toy Story 5' and Live-Action 'Moana' for 2026 Release Dates

Streaming right now: What to watch and the best of what&#x27;s new on Netflix, Hulu, HBO Max, Disney+, Apple TV+, Peacock, and Paramount+.

5 must-watch movies & TV shows streaming right now

Star Wars: The Dawn of the Jedi has enlisted Beau Willimon to work with James Mangold on his prequel movie set 250,000 years in the past.

'Star Wars': Beau Willimon to Co-Write James Mangold's Movie (Exclusive)

Jared Leto&#x27;s &#x27;Tron: Ares&#x27; also gets a date amid a slew of changes.

Disney Sets 'Toy Story 5,' 'Mandalorian & Grogu' for 2026, Delays Live-Action 'Moana' by a Year

Movio

movio

In a highly controversial move, Kevin Rose announced yesterday that Moonbirds and Oddities are moving to the CC0 public license. https://twitter.com/kevinrose/status/1555262099093200896 No Rights Reserved CC0 enables creators and owners of copyright-protected content to waive those rights and thereby place the creations in the public domain. This would allow anyone to freely use the works for any purpose without restriction under copyright law. In simple terms, this would mean that anyone can freely use the Moonbirds and Oddities art in any way they see fit. This effectively legitimizes all the derivative NFT collections that have been floating around including AI and 3D Moonbirds. https://twitter.com/kevinrose/status/1555262101320392704?ref_src=twsrc%5Etfw%7Ctwcamp%5Etweetembed%7Ctwterm%5E1555262102410915840%7Ctwgr%5E84cf975d4cdc9e694846be27605622741337ed00%7Ctwcon%5Es2_&ref_url=https%3A%2F%2Fhackernoon.com%2Fwhy-did-moonbirds-move-all-user-generated-art-to-the-creative-commons-license The controversy While the move to CC0 seems to honor the basic principles of Web 3, this major shift has not been received positively by everyone. Many holders of these expensive JPEGs were caught off guard by the CC0 announcement and felt a discussion was warranted for such a massive change. Moreover, many NFT holders who bought into the project believed they had IP rights to their JPEGs and some even had plans to use their specific bird or oddity to launch brands in the future. Building a business using these JPEGs as branding now becomes complicated as the art is effectively in the public domain and available for anyone to use and repurpose. The upsides In the spirit of Web 3, true ownership will no longer be determined by a central entity. Provenance will now come from the blockchain itself as God intended. Once something is in the public domain, that right is irrevocable. There are no take backsies here even if Proof were to be sold off and managed by a different organization down the line. The move to CC0 has massive implications regarding the Moonbirds and the Oddities IP. I understand why people are upset. If I had plans to launch a brand using the Moonbirds IP, I would be frustrated as well. However, I have always looked at these illiquid JPEGs as just collectibles whose true utility comes from social signaling. To that end, the provenance that comes from the blockchain should be the only thing that matters. Although people can copy and repurpose your JPEG, what should truly matter is that you own the original. https://twitter.com/kevinrose/status/1555262109969092614 Thank you for reading. Check out my unfiltered thoughts on and follow my career . Twitter on LinkedIn

Why Did Moonbirds Move All User Generated Art To the Creative Commons License? 

¿Por qué Moonbirds movió todo el arte generado por los usuarios a la licencia Creative Commons?

Pourquoi Moonbirds a-t-il déplacé toutes les illustrations générées par les utilisateurs vers la licence Creative Commons ?

मूनबर्ड्स ने सभी उपयोगकर्ता जनित कला को क्रिएटिव कॉमन्स लाइसेंस में क्यों स्थानांतरित किया?

Moonbirds がすべてのユーザー生成アートをクリエイティブ コモンズ ライセンスに移行したのはなぜですか?

Por que os Moonbirds moveram todas as artes geradas pelo usuário para a licença Creative Commons?

Tại sao Moonbirds chuyển tất cả tác phẩm do người dùng sáng tạo sang giấy phép Creative Commons?

为什么 Moonbirds 将所有用户生成的艺术转移到知识共享许可中？

Authors:
(1) Yuwei Guo, The Chinese University of Hong Kong;
(2) Ceyuan Yang, Shanghai Artificial Intelligence Laboratory with Corresponding Author;
(3) Anyi Rao, Stanford University;
(4) Zhengyang Liang, Shanghai Artificial Intelligence Laboratory;
(5) Yaohui Wang, Shanghai Artificial Intelligence Laboratory;
(6) Yu Qiao, Shanghai Artificial Intelligence Laboratory;
(7) Maneesh Agrawala, Stanford University;
(8) Dahua Lin, Shanghai Artificial Intelligence Laboratory;
(9) Bo Dai, The Chinese University of Hong Kong and The Chinese University of Hong Kong. Table of Links Abstract and 1 Introduction 2 Work Related 3 Preliminary AnimateDiff 4.1 Alleviate Negative Effects from Training Data with Domain Adapter 4.2 Learn Motion Priors with Motion Module 4.3 Adapt to New Motion Patterns with MotionLora 4.4 AnimateDiff in Practice 5 Experiments and 5.1 Qualitative Results 5.2 Qualitative Comparison 5.3 Ablative Study 5.4 Controllable Generation 6 Conclusion 7 Ethics Statement 8 Reproducibility Statement, Acknowledgement and References 6 CONCLUSION In this paper, we present AnimateDiff, a practical pipeline directly turning personalized text-toimage (T2I) models for animation generation once and for all, without compromising quality or losing pre-learned domain knowledge. To accomplish this, we design three component modules in AnimateDiff to learn meaningful motion priors while alleviating visual quality degradation and enabling motion personalization with a lightweight fine-tuning technique named MotionLoRA. Once trained, our motion module can be integrated into other personalized T2Is to generate animated images with natural and coherent motions while remaining faithful to the personalized domain. Extensive evaluation with various personalized T2I models also validates the effectiveness and generalizability of our AnimateDiff and MotionLoRA. Furthermore, we demonstrate the compatibility of our method with existing content-controlling approaches, enabling controllable generation without incurring additional training costs. Overall, AnimateDiff provides an effective baseline for personalized animation and holds significant potential for a wide range of applications. This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Yuwei Guo, The Chinese University of Hong Kong; (2) Ceyuan Yang, Shanghai Artificial Intelligence Laboratory with Corresponding Author; (3) Anyi Rao, Stanford University; (4) Zhengyang Liang, Shanghai Artificial Intelligence Laboratory; (5) Yaohui Wang, Shanghai Artificial Intelligence Laboratory; (6) Yu Qiao, Shanghai Artificial Intelligence Laboratory; (7) Maneesh Agrawala, Stanford University; (8) Dahua Lin, Shanghai Artificial Intelligence Laboratory; (9) Bo Dai, The Chinese University of Hong Kong and The Chinese University of Hong Kong. Authors: Authors: (1) Yuwei Guo, The Chinese University of Hong Kong; (2) Ceyuan Yang, Shanghai Artificial Intelligence Laboratory with Corresponding Author; (3) Anyi Rao, Stanford University; (4) Zhengyang Liang, Shanghai Artificial Intelligence Laboratory; (5) Yaohui Wang, Shanghai Artificial Intelligence Laboratory; (6) Yu Qiao, Shanghai Artificial Intelligence Laboratory; (7) Maneesh Agrawala, Stanford University; (8) Dahua Lin, Shanghai Artificial Intelligence Laboratory; (9) Bo Dai, The Chinese University of Hong Kong and The Chinese University of Hong Kong. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Work Related 2 Work Related 3 Preliminary 3 Preliminary AnimateDiff AnimateDiff 4.1 Alleviate Negative Effects from Training Data with Domain Adapter 4.1 Alleviate Negative Effects from Training Data with Domain Adapter 4.2 Learn Motion Priors with Motion Module 4.2 Learn Motion Priors with Motion Module 4.3 Adapt to New Motion Patterns with MotionLora 4.3 Adapt to New Motion Patterns with MotionLora 4.4 AnimateDiff in Practice 4.4 AnimateDiff in Practice 5 Experiments and 5.1 Qualitative Results 5 Experiments and 5.1 Qualitative Results 5.2 Qualitative Comparison 5.2 Qualitative Comparison 5.3 Ablative Study 5.3 Ablative Study 5.4 Controllable Generation 5.4 Controllable Generation 6 Conclusion 6 Conclusion 7 Ethics Statement 7 Ethics Statement 8 Reproducibility Statement, Acknowledgement and References 8 Reproducibility Statement, Acknowledgement and References 6 CONCLUSION In this paper, we present AnimateDiff, a practical pipeline directly turning personalized text-toimage (T2I) models for animation generation once and for all, without compromising quality or losing pre-learned domain knowledge. To accomplish this, we design three component modules in AnimateDiff to learn meaningful motion priors while alleviating visual quality degradation and enabling motion personalization with a lightweight fine-tuning technique named MotionLoRA. Once trained, our motion module can be integrated into other personalized T2Is to generate animated images with natural and coherent motions while remaining faithful to the personalized domain. Extensive evaluation with various personalized T2I models also validates the effectiveness and generalizability of our AnimateDiff and MotionLoRA. Furthermore, we demonstrate the compatibility of our method with existing content-controlling approaches, enabling controllable generation without incurring additional training costs. Overall, AnimateDiff provides an effective baseline for personalized animation and holds significant potential for a wide range of applications. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv

How AnimateDiff Transforms T2I Models into High-Quality Animation Generators with MotionLORA

Authors:
(1) Yuwei Guo, The Chinese University of Hong Kong;
(2) Ceyuan Yang, Shanghai Artificial Intelligence Laboratory with Corresponding Author;
(3) Anyi Rao, Stanford University;
(4) Zhengyang Liang, Shanghai Artificial Intelligence Laboratory;
(5) Yaohui Wang, Shanghai Artificial Intelligence Laboratory;
(6) Yu Qiao, Shanghai Artificial Intelligence Laboratory;
(7) Maneesh Agrawala, Stanford University;
(8) Dahua Lin, Shanghai Artificial Intelligence Laboratory;
(9) Bo Dai, The Chinese University of Hong Kong and The Chinese University of Hong Kong. Table of Links Abstract and 1 Introduction 2 Work Related 3 Preliminary AnimateDiff 4.1 Alleviate Negative Effects from Training Data with Domain Adapter 4.2 Learn Motion Priors with Motion Module 4.3 Adapt to New Motion Patterns with MotionLora 4.4 AnimateDiff in Practice 5 Experiments and 5.1 Qualitative Results 5.2 Qualitative Comparison 5.3 Ablative Study 5.4 Controllable Generation 6 Conclusion 7 Ethics Statement 8 Reproducibility Statement, Acknowledgement and References 5.4 CONTROLLABLE GENERATION. The separated learning of visual content and motion priors in AnimateDiff enables the direct application of existing content control approaches for controllable generation. To demonstrate this capability, we combined AnimateDiff with ControlNet (Zhang et al., 2023) to control the generation with extracted depth map sequence. In contrast to recent video editing techniques (Ceylan et al., 2023; Wang et al., 2023a) that employ DDIM (Song et al., 2020) inversion to obtain smoothed latent sequences, we generate animations from randomly sampled noise. As illustrated in Figure 8, our results exhibit meticulous motion details (such as hair and facial expressions) and high visual quality. This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Yuwei Guo, The Chinese University of Hong Kong; (2) Ceyuan Yang, Shanghai Artificial Intelligence Laboratory with Corresponding Author; (3) Anyi Rao, Stanford University; (4) Zhengyang Liang, Shanghai Artificial Intelligence Laboratory; (5) Yaohui Wang, Shanghai Artificial Intelligence Laboratory; (6) Yu Qiao, Shanghai Artificial Intelligence Laboratory; (7) Maneesh Agrawala, Stanford University; (8) Dahua Lin, Shanghai Artificial Intelligence Laboratory; (9) Bo Dai, The Chinese University of Hong Kong and The Chinese University of Hong Kong. Authors: Authors: (1) Yuwei Guo, The Chinese University of Hong Kong; (2) Ceyuan Yang, Shanghai Artificial Intelligence Laboratory with Corresponding Author; (3) Anyi Rao, Stanford University; (4) Zhengyang Liang, Shanghai Artificial Intelligence Laboratory; (5) Yaohui Wang, Shanghai Artificial Intelligence Laboratory; (6) Yu Qiao, Shanghai Artificial Intelligence Laboratory; (7) Maneesh Agrawala, Stanford University; (8) Dahua Lin, Shanghai Artificial Intelligence Laboratory; (9) Bo Dai, The Chinese University of Hong Kong and The Chinese University of Hong Kong. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Work Related 2 Work Related 3 Preliminary 3 Preliminary AnimateDiff AnimateDiff 4.1 Alleviate Negative Effects from Training Data with Domain Adapter 4.1 Alleviate Negative Effects from Training Data with Domain Adapter 4.2 Learn Motion Priors with Motion Module 4.2 Learn Motion Priors with Motion Module 4.3 Adapt to New Motion Patterns with MotionLora 4.3 Adapt to New Motion Patterns with MotionLora 4.4 AnimateDiff in Practice 4.4 AnimateDiff in Practice 5 Experiments and 5.1 Qualitative Results 5 Experiments and 5.1 Qualitative Results 5.2 Qualitative Comparison 5.2 Qualitative Comparison 5.3 Ablative Study 5.3 Ablative Study 5.4 Controllable Generation 5.4 Controllable Generation 6 Conclusion 6 Conclusion 7 Ethics Statement 7 Ethics Statement 8 Reproducibility Statement, Acknowledgement and References 8 Reproducibility Statement, Acknowledgement and References 5.4 CONTROLLABLE GENERATION. The separated learning of visual content and motion priors in AnimateDiff enables the direct application of existing content control approaches for controllable generation. To demonstrate this capability, we combined AnimateDiff with ControlNet (Zhang et al., 2023) to control the generation with extracted depth map sequence. In contrast to recent video editing techniques (Ceylan et al., 2023; Wang et al., 2023a) that employ DDIM (Song et al., 2020) inversion to obtain smoothed latent sequences, we generate animations from randomly sampled noise. As illustrated in Figure 8, our results exhibit meticulous motion details (such as hair and facial expressions) and high visual quality. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv

AnimateDiff Combines with ControlNet for Precise Motion Control and High-Quality Video Generation

Authors:
(1) Yuwei Guo, The Chinese University of Hong Kong;
(2) Ceyuan Yang, Shanghai Artificial Intelligence Laboratory with Corresponding Author;
(3) Anyi Rao, Stanford University;
(4) Zhengyang Liang, Shanghai Artificial Intelligence Laboratory;
(5) Yaohui Wang, Shanghai Artificial Intelligence Laboratory;
(6) Yu Qiao, Shanghai Artificial Intelligence Laboratory;
(7) Maneesh Agrawala, Stanford University;
(8) Dahua Lin, Shanghai Artificial Intelligence Laboratory;
(9) Bo Dai, The Chinese University of Hong Kong and The Chinese University of Hong Kong. Table of Links Abstract and 1 Introduction 2 Work Related 3 Preliminary AnimateDiff 4.1 Alleviate Negative Effects from Training Data with Domain Adapter 4.2 Learn Motion Priors with Motion Module 4.3 Adapt to New Motion Patterns with MotionLora 4.4 AnimateDiff in Practice 5 Experiments and 5.1 Qualitative Results 5.2 Qualitative Comparison 5.3 Ablative Study 5.4 Controllable Generation 6 Conclusion 7 Ethics Statement 8 Reproducibility Statement, Acknowledgement and References 5.3 ABLATIVE STUDY Domain adapter. To investigate the impact of the domain adapter in AnimateDiff, we conducted a study by adjusting the scaler in the adapter layers during inference, ranging from 1 (full impact) to 0 (complete removal). As illustrated in Figure 6, as the scaler of the adapter decreases, there is an improvement in overall visual quality, accompanied by a reduction in the visual content distribution learned from the video dataset (the watermark in the case of WebVid (Bain et al., 2021)). These results indicate the successful role of the domain adapter in enhancing the visual quality of AnimateDiff by alleviating the motion module from learning the visual distribution gap. Motion module design. We compare our motion module design of the temporal Transformer with its full convolution counterpart, which is motivated by the fact that both designs are widely employed in recent works on video generation. We replace the temporal attention with 1D temporal convolution and ensured that the two model parameters were closely aligned. As depicted in supplementary materials, the convolutional motion module aligns all frames to be identical but does not incorporate any motion compared to the Transformer architecture. Efficiency of MotionLoRA. The efficiency of MotionLoRA in AnimateDiff was examined in terms of parameter efficiency and data efficiency. Parameter efficiency is crucial for efficient model training and sharing among users, while data efficiency is essential for real-world applications where collecting an adequate number of reference videos for specific motion patterns may be challenging. To investigate these aspects, we trained multiple MotionLoRA models with varying parameter scales and reference video quantities. In Fig. 7, the first two samples demonstrate that MotionLoRA is capable of learning new camera motions (e.g., zoom-in) with a small parameter scale while maintaining comparable motion quality. Furthermore, even with a modest number of reference videos (e.g., N = 50), the model successfully learns the desired motion patterns. However, when the number of reference videos is excessively limited (e.g., N = 5), significant degradation in quality is observed, suggesting that MotionLoRA encounters difficulties in learning shared motion patterns and instead relies on capturing texture information from the reference videos. This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Yuwei Guo, The Chinese University of Hong Kong; (2) Ceyuan Yang, Shanghai Artificial Intelligence Laboratory with Corresponding Author; (3) Anyi Rao, Stanford University; (4) Zhengyang Liang, Shanghai Artificial Intelligence Laboratory; (5) Yaohui Wang, Shanghai Artificial Intelligence Laboratory; (6) Yu Qiao, Shanghai Artificial Intelligence Laboratory; (7) Maneesh Agrawala, Stanford University; (8) Dahua Lin, Shanghai Artificial Intelligence Laboratory; (9) Bo Dai, The Chinese University of Hong Kong and The Chinese University of Hong Kong. Authors: Authors: (1) Yuwei Guo, The Chinese University of Hong Kong; (2) Ceyuan Yang, Shanghai Artificial Intelligence Laboratory with Corresponding Author; (3) Anyi Rao, Stanford University; (4) Zhengyang Liang, Shanghai Artificial Intelligence Laboratory; (5) Yaohui Wang, Shanghai Artificial Intelligence Laboratory; (6) Yu Qiao, Shanghai Artificial Intelligence Laboratory; (7) Maneesh Agrawala, Stanford University; (8) Dahua Lin, Shanghai Artificial Intelligence Laboratory; (9) Bo Dai, The Chinese University of Hong Kong and The Chinese University of Hong Kong. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Work Related 2 Work Related 3 Preliminary 3 Preliminary AnimateDiff AnimateDiff 4.1 Alleviate Negative Effects from Training Data with Domain Adapter 4.1 Alleviate Negative Effects from Training Data with Domain Adapter 4.2 Learn Motion Priors with Motion Module 4.2 Learn Motion Priors with Motion Module 4.3 Adapt to New Motion Patterns with MotionLora 4.3 Adapt to New Motion Patterns with MotionLora 4.4 AnimateDiff in Practice 4.4 AnimateDiff in Practice 5 Experiments and 5.1 Qualitative Results 5 Experiments and 5.1 Qualitative Results 5.2 Qualitative Comparison 5.2 Qualitative Comparison 5.3 Ablative Study 5.3 Ablative Study 5.4 Controllable Generation 5.4 Controllable Generation 6 Conclusion 6 Conclusion 7 Ethics Statement 7 Ethics Statement 8 Reproducibility Statement, Acknowledgement and References 8 Reproducibility Statement, Acknowledgement and References 5.3 ABLATIVE STUDY Domain adapter. To investigate the impact of the domain adapter in AnimateDiff, we conducted a study by adjusting the scaler in the adapter layers during inference, ranging from 1 (full impact) to 0 (complete removal). As illustrated in Figure 6, as the scaler of the adapter decreases, there is an improvement in overall visual quality, accompanied by a reduction in the visual content distribution learned from the video dataset (the watermark in the case of WebVid (Bain et al., 2021)). These results indicate the successful role of the domain adapter in enhancing the visual quality of AnimateDiff by alleviating the motion module from learning the visual distribution gap. Domain adapter. Motion module design. We compare our motion module design of the temporal Transformer with its full convolution counterpart, which is motivated by the fact that both designs are widely employed in recent works on video generation. We replace the temporal attention with 1D temporal convolution and ensured that the two model parameters were closely aligned. As depicted in supplementary materials, the convolutional motion module aligns all frames to be identical but does not incorporate any motion compared to the Transformer architecture. Motion module design. Efficiency of MotionLoRA. The efficiency of MotionLoRA in AnimateDiff was examined in terms of parameter efficiency and data efficiency. Parameter efficiency is crucial for efficient model training and sharing among users, while data efficiency is essential for real-world applications where collecting an adequate number of reference videos for specific motion patterns may be challenging. To investigate these aspects, we trained multiple MotionLoRA models with varying parameter scales and reference video quantities. In Fig. 7, the first two samples demonstrate that MotionLoRA is capable of learning new camera motions (e.g., zoom-in) with a small parameter scale while maintaining comparable motion quality. Furthermore, even with a modest number of reference videos (e.g., N = 50), the model successfully learns the desired motion patterns. However, when the number of reference videos is excessively limited (e.g., N = 5), significant degradation in quality is observed, suggesting that MotionLoRA encounters difficulties in learning shared motion patterns and instead relies on capturing texture information from the reference videos. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv

Ablative Study on Domain Adapter, Motion Module Design, and MotionLoRA Efficiency

Authors:
(1) Yuwei Guo, The Chinese University of Hong Kong;
(2) Ceyuan Yang, Shanghai Artificial Intelligence Laboratory with Corresponding Author;
(3) Anyi Rao, Stanford University;
(4) Zhengyang Liang, Shanghai Artificial Intelligence Laboratory;
(5) Yaohui Wang, Shanghai Artificial Intelligence Laboratory;
(6) Yu Qiao, Shanghai Artificial Intelligence Laboratory;
(7) Maneesh Agrawala, Stanford University;
(8) Dahua Lin, Shanghai Artificial Intelligence Laboratory;
(9) Bo Dai, The Chinese University of Hong Kong and The Chinese University of Hong Kong. Table of Links Abstract and 1 Introduction 2 Work Related 3 Preliminary AnimateDiff 4.1 Alleviate Negative Effects from Training Data with Domain Adapter 4.2 Learn Motion Priors with Motion Module 4.3 Adapt to New Motion Patterns with MotionLora 4.4 AnimateDiff in Practice 5 Experiments and 5.1 Qualitative Results 5.2 Qualitative Comparison 5.3 Ablative Study 5.4 Controllable Generation 6 Conclusion 7 Ethics Statement 8 Reproducibility Statement, Acknowledgement and References 5 EXPERIMENTS We implement AnimateDiff upon Stable Diffusion V1.5 and train motion module using the WebVid10M (Bain et al., 2021) dataset. Detailed configurations can be found in supplementary materials. 5.1 QUALITATIVE RESULTS Evaluate on community models. We evaluated the AnimateDiff with a diverse set of representative personalized T2Is collected from Civitai (2022). These personalized T2Is encompass a wide range of domains, thus serving as a comprehensive benchmark. Since personalized domains in these T2Is only respond to certain “trigger words”, we abstain from using common text prompts but refer to the model homepage to construct the evaluation prompts. In Fig. 4, we show eight qualitative results of AnimateDiff. Each sample corresponds to a distinct personalized T2I. In the second row of Figure 1, we present the outcomes obtained by integrating AnimateDiff with MotionLoRA to achieve shot type controls. The last two samples exhibit the composition capability of MotionLoRA, achieved by linearly combining the individually trained weights. Compare with baselines. In the absence of existing methods specifically designed for animating personalized T2Is, we compare our method with two recent works in video generation that can be adapted for this task: 1) Text2Video-Zero (Khachatryan et al., 2023) and 2) Tune-a-Video (Wu 7 et al., 2023). We also compare AnimateDiff with two commercial tools: 3) Gen-2 (2023) for textto-video generation, and 4) Pika Labs (2023) for image animation. The results are shown in Fig. 5. This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Yuwei Guo, The Chinese University of Hong Kong; (2) Ceyuan Yang, Shanghai Artificial Intelligence Laboratory with Corresponding Author; (3) Anyi Rao, Stanford University; (4) Zhengyang Liang, Shanghai Artificial Intelligence Laboratory; (5) Yaohui Wang, Shanghai Artificial Intelligence Laboratory; (6) Yu Qiao, Shanghai Artificial Intelligence Laboratory; (7) Maneesh Agrawala, Stanford University; (8) Dahua Lin, Shanghai Artificial Intelligence Laboratory; (9) Bo Dai, The Chinese University of Hong Kong and The Chinese University of Hong Kong. Authors: Authors: (1) Yuwei Guo, The Chinese University of Hong Kong; (2) Ceyuan Yang, Shanghai Artificial Intelligence Laboratory with Corresponding Author; (3) Anyi Rao, Stanford University; (4) Zhengyang Liang, Shanghai Artificial Intelligence Laboratory; (5) Yaohui Wang, Shanghai Artificial Intelligence Laboratory; (6) Yu Qiao, Shanghai Artificial Intelligence Laboratory; (7) Maneesh Agrawala, Stanford University; (8) Dahua Lin, Shanghai Artificial Intelligence Laboratory; (9) Bo Dai, The Chinese University of Hong Kong and The Chinese University of Hong Kong. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Work Related 2 Work Related 3 Preliminary 3 Preliminary AnimateDiff AnimateDiff 4.1 Alleviate Negative Effects from Training Data with Domain Adapter 4.1 Alleviate Negative Effects from Training Data with Domain Adapter 4.2 Learn Motion Priors with Motion Module 4.2 Learn Motion Priors with Motion Module 4.3 Adapt to New Motion Patterns with MotionLora 4.3 Adapt to New Motion Patterns with MotionLora 4.4 AnimateDiff in Practice 4.4 AnimateDiff in Practice 5 Experiments and 5.1 Qualitative Results 5 Experiments and 5.1 Qualitative Results 5.2 Qualitative Comparison 5.2 Qualitative Comparison 5.3 Ablative Study 5.3 Ablative Study 5.4 Controllable Generation 5.4 Controllable Generation 6 Conclusion 6 Conclusion 7 Ethics Statement 7 Ethics Statement 8 Reproducibility Statement, Acknowledgement and References 8 Reproducibility Statement, Acknowledgement and References 5 EXPERIMENTS We implement AnimateDiff upon Stable Diffusion V1.5 and train motion module using the WebVid10M (Bain et al., 2021) dataset. Detailed configurations can be found in supplementary materials. 5.1 QUALITATIVE RESULTS Evaluate on community models. We evaluated the AnimateDiff with a diverse set of representative personalized T2Is collected from Civitai (2022). These personalized T2Is encompass a wide range of domains, thus serving as a comprehensive benchmark. Since personalized domains in these T2Is only respond to certain “trigger words”, we abstain from using common text prompts but refer to the model homepage to construct the evaluation prompts. In Fig. 4, we show eight qualitative results of AnimateDiff. Each sample corresponds to a distinct personalized T2I. In the second row of Figure 1, we present the outcomes obtained by integrating AnimateDiff with MotionLoRA to achieve shot type controls. The last two samples exhibit the composition capability of MotionLoRA, achieved by linearly combining the individually trained weights. Compare with baselines. In the absence of existing methods specifically designed for animating personalized T2Is, we compare our method with two recent works in video generation that can be adapted for this task: 1) Text2Video-Zero (Khachatryan et al., 2023) and 2) Tune-a-Video (Wu 7 et al., 2023). We also compare AnimateDiff with two commercial tools: 3) Gen-2 (2023) for textto-video generation, and 4) Pika Labs (2023) for image animation. The results are shown in Fig. 5. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv

How AnimateDiff Brings Personalized T2Is to Life with Efficient Motion Modeling

Authors:
(1) Yuwei Guo, The Chinese University of Hong Kong;
(2) Ceyuan Yang, Shanghai Artificial Intelligence Laboratory with Corresponding Author;
(3) Anyi Rao, Stanford University;
(4) Zhengyang Liang, Shanghai Artificial Intelligence Laboratory;
(5) Yaohui Wang, Shanghai Artificial Intelligence Laboratory;
(6) Yu Qiao, Shanghai Artificial Intelligence Laboratory;
(7) Maneesh Agrawala, Stanford University;
(8) Dahua Lin, Shanghai Artificial Intelligence Laboratory;
(9) Bo Dai, The Chinese University of Hong Kong and The Chinese University of Hong Kong. Table of Links Abstract and 1 Introduction 2 Work Related 3 Preliminary AnimateDiff 4.1 Alleviate Negative Effects from Training Data with Domain Adapter 4.2 Learn Motion Priors with Motion Module 4.3 Adapt to New Motion Patterns with MotionLora 4.4 AnimateDiff in Practice 5 Experiments and 5.1 Qualitative Results 5.2 Qualitative Comparison 5.3 Ablative Study 5.4 Controllable Generation 6 Conclusion 7 Ethics Statement 8 Reproducibility Statement, Acknowledgement and References 4.3 ADAPT TO NEW MOTION PATTERNS WITH MOTIONLORA While the pre-trained motion module captures general motion priors, a question arises when we need to effectively adapt it to new motion patterns such as camera zooming, panning and rolling, etc., with a small number of reference videos and training iterations. Such efficiency is essential for users who cannot afford expensive pre-training costs but would like to fine-tune the motion module for specific effects. Here comes the last stage of AnimateDiff, also dubbed as MotionLoRA (Fig. 3), an efficient fine-tuning approach for motion personalization. Considering the architecture of the motion module and the limited number of reference videos, we add LoRA layers to the self-attention layers of the motion module in the inflated model described in Sec. 4.2, then train these LoRA layers on the reference videos of new motion patterns. We experiment with several shot types and get the reference videos via rule-based data augmentation. For instance, to get videos with zooming effects, we augment the videos by gradually reducing (zoom-in) or enlarging (zoom-out) the cropping area of video frames along the temporal axis. We demonstrate that our MotionLoRA can achieve promising results even with as few as 20 ∼ 50 reference videos, 2,000 training iterations (around 1 ∼ 2 hours) as well as about 30M storage space, enabling efficient model tuning and sharing among users. Benefited by the low-rank property, MotionLoRA also has the composition capability. Namely, individually trained MotionLoRA models can be combined to achieve composed motion effects at inference time. This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Yuwei Guo, The Chinese University of Hong Kong; (2) Ceyuan Yang, Shanghai Artificial Intelligence Laboratory with Corresponding Author; (3) Anyi Rao, Stanford University; (4) Zhengyang Liang, Shanghai Artificial Intelligence Laboratory; (5) Yaohui Wang, Shanghai Artificial Intelligence Laboratory; (6) Yu Qiao, Shanghai Artificial Intelligence Laboratory; (7) Maneesh Agrawala, Stanford University; (8) Dahua Lin, Shanghai Artificial Intelligence Laboratory; (9) Bo Dai, The Chinese University of Hong Kong and The Chinese University of Hong Kong. Authors: Authors: (1) Yuwei Guo, The Chinese University of Hong Kong; (2) Ceyuan Yang, Shanghai Artificial Intelligence Laboratory with Corresponding Author; (3) Anyi Rao, Stanford University; (4) Zhengyang Liang, Shanghai Artificial Intelligence Laboratory; (5) Yaohui Wang, Shanghai Artificial Intelligence Laboratory; (6) Yu Qiao, Shanghai Artificial Intelligence Laboratory; (7) Maneesh Agrawala, Stanford University; (8) Dahua Lin, Shanghai Artificial Intelligence Laboratory; (9) Bo Dai, The Chinese University of Hong Kong and The Chinese University of Hong Kong. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Work Related 2 Work Related 3 Preliminary 3 Preliminary AnimateDiff AnimateDiff 4.1 Alleviate Negative Effects from Training Data with Domain Adapter 4.1 Alleviate Negative Effects from Training Data with Domain Adapter 4.2 Learn Motion Priors with Motion Module 4.2 Learn Motion Priors with Motion Module 4.3 Adapt to New Motion Patterns with MotionLora 4.3 Adapt to New Motion Patterns with MotionLora 4.4 AnimateDiff in Practice 4.4 AnimateDiff in Practice 5 Experiments and 5.1 Qualitative Results 5 Experiments and 5.1 Qualitative Results 5.2 Qualitative Comparison 5.2 Qualitative Comparison 5.3 Ablative Study 5.3 Ablative Study 5.4 Controllable Generation 5.4 Controllable Generation 6 Conclusion 6 Conclusion 7 Ethics Statement 7 Ethics Statement 8 Reproducibility Statement, Acknowledgement and References 8 Reproducibility Statement, Acknowledgement and References 4.3 ADAPT TO NEW MOTION PATTERNS WITH MOTIONLORA While the pre-trained motion module captures general motion priors, a question arises when we need to effectively adapt it to new motion patterns such as camera zooming, panning and rolling, etc., with a small number of reference videos and training iterations. Such efficiency is essential for users who cannot afford expensive pre-training costs but would like to fine-tune the motion module for specific effects. Here comes the last stage of AnimateDiff, also dubbed as MotionLoRA (Fig. 3), an efficient fine-tuning approach for motion personalization. Considering the architecture of the motion module and the limited number of reference videos, we add LoRA layers to the self-attention layers of the motion module in the inflated model described in Sec. 4.2, then train these LoRA layers on the reference videos of new motion patterns. We experiment with several shot types and get the reference videos via rule-based data augmentation. For instance, to get videos with zooming effects, we augment the videos by gradually reducing (zoom-in) or enlarging (zoom-out) the cropping area of video frames along the temporal axis. We demonstrate that our MotionLoRA can achieve promising results even with as few as 20 ∼ 50 reference videos, 2,000 training iterations (around 1 ∼ 2 hours) as well as about 30M storage space, enabling efficient model tuning and sharing among users. Benefited by the low-rank property, MotionLoRA also has the composition capability. Namely, individually trained MotionLoRA models can be combined to achieve composed motion effects at inference time. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv

Adapting Motion Patterns Efficiently with MotionLoRA in AnimateDiff

Authors:
(1) Yuwei Guo, The Chinese University of Hong Kong;
(2) Ceyuan Yang, Shanghai Artificial Intelligence Laboratory with Corresponding Author;
(3) Anyi Rao, Stanford University;
(4) Zhengyang Liang, Shanghai Artificial Intelligence Laboratory;
(5) Yaohui Wang, Shanghai Artificial Intelligence Laboratory;
(6) Yu Qiao, Shanghai Artificial Intelligence Laboratory;
(7) Maneesh Agrawala, Stanford University;
(8) Dahua Lin, Shanghai Artificial Intelligence Laboratory;
(9) Bo Dai, The Chinese University of Hong Kong and The Chinese University of Hong Kong. Table of Links Abstract and 1 Introduction 2 Work Related 3 Preliminary AnimateDiff 4.1 Alleviate Negative Effects from Training Data with Domain Adapter 4.2 Learn Motion Priors with Motion Module 4.3 Adapt to New Motion Patterns with MotionLora 4.4 AnimateDiff in Practice 5 Experiments and 5.1 Qualitative Results 5.2 Qualitative Comparison 5.3 Ablative Study 5.4 Controllable Generation 6 Conclusion 7 Ethics Statement 8 Reproducibility Statement, Acknowledgement and References 4.2 LEARN MOTION PRIORS WITH MOTION MODULE To model motion dynamics along the temporal dimension on top of a pre-trained T2I, we must 1) inflate the 2-dimensional diffusion model to deal with 3-dimensional video data and 2) design a sub-module to enable efficient information exchange along the temporal axis. This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Yuwei Guo, The Chinese University of Hong Kong; (2) Ceyuan Yang, Shanghai Artificial Intelligence Laboratory with Corresponding Author; (3) Anyi Rao, Stanford University; (4) Zhengyang Liang, Shanghai Artificial Intelligence Laboratory; (5) Yaohui Wang, Shanghai Artificial Intelligence Laboratory; (6) Yu Qiao, Shanghai Artificial Intelligence Laboratory; (7) Maneesh Agrawala, Stanford University; (8) Dahua Lin, Shanghai Artificial Intelligence Laboratory; (9) Bo Dai, The Chinese University of Hong Kong and The Chinese University of Hong Kong. Authors: Authors: (1) Yuwei Guo, The Chinese University of Hong Kong; (2) Ceyuan Yang, Shanghai Artificial Intelligence Laboratory with Corresponding Author; (3) Anyi Rao, Stanford University; (4) Zhengyang Liang, Shanghai Artificial Intelligence Laboratory; (5) Yaohui Wang, Shanghai Artificial Intelligence Laboratory; (6) Yu Qiao, Shanghai Artificial Intelligence Laboratory; (7) Maneesh Agrawala, Stanford University; (8) Dahua Lin, Shanghai Artificial Intelligence Laboratory; (9) Bo Dai, The Chinese University of Hong Kong and The Chinese University of Hong Kong. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Work Related 2 Work Related 3 Preliminary 3 Preliminary AnimateDiff AnimateDiff 4.1 Alleviate Negative Effects from Training Data with Domain Adapter 4.1 Alleviate Negative Effects from Training Data with Domain Adapter 4.2 Learn Motion Priors with Motion Module 4.2 Learn Motion Priors with Motion Module 4.3 Adapt to New Motion Patterns with MotionLora 4.3 Adapt to New Motion Patterns with MotionLora 4.4 AnimateDiff in Practice 4.4 AnimateDiff in Practice 5 Experiments and 5.1 Qualitative Results 5 Experiments and 5.1 Qualitative Results 5.2 Qualitative Comparison 5.2 Qualitative Comparison 5.3 Ablative Study 5.3 Ablative Study 5.4 Controllable Generation 5.4 Controllable Generation 6 Conclusion 6 Conclusion 7 Ethics Statement 7 Ethics Statement 8 Reproducibility Statement, Acknowledgement and References 8 Reproducibility Statement, Acknowledgement and References 4.2 LEARN MOTION PRIORS WITH MOTION MODULE To model motion dynamics along the temporal dimension on top of a pre-trained T2I, we must 1) inflate the 2-dimensional diffusion model to deal with 3-dimensional video data and 2) design a sub-module to enable efficient information exchange along the temporal axis. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv

Mastering Motion Dynamics in Animation with Temporal Transformers

Have you ever watched a movie on YouTube? Oh, yes you have! If you’re lucky and stubborn enough, you can stumble upon one that’s so old and well-known that everybody stopped caring about it. I mean when it comes to copyrights and stuff. Did you know that “Conan The Barbarian” is a big deal in GREECE? It’s not because of Arnold Schwarzenegger but Basil Poledouris. It’s not because of the movie itself but the timeless soundtrack. So, if you want to have royal treatment with the locals while spending a summer in Greece, try wearing a T-shirt with a Conan movie poster instead of “This Is Sparta” (no disrespect to Gerard Butler), and listening to Prologue/Anvil of Crom all day long, so everyone around can hear it. Wait and see what happens. There’s “Conan The Barbarian” with Greek subtitles available on YouTube at any given moment, but that’s not what I’m writing about. Give It To Me, YouTube Baby I’m talking about the entire collections of Marvel Avengers and DCU movies in the best resolution imaginable free to watch on YouTube. If you aren’t impressed enough, then how about Venom 3: The Last Dance, Deadpool 3, or Dune 2, just off the top of my head? There’s an invasion of YouTube channels with movies you couldn’t believe are available for free in top-resolution quality. There’s a catch, though. You won’t know what you’re about to watch until you start watching. All videos have identical “cryptic” descriptions: “BLOCKBUSTER Movie 2024.” It’s like playing Russian roulette but in a good and enjoyable way. Also, I’m just guessing because I’m not a YouTube expert, all these channels have a handful of initial “innocent” videos of ordinary people doing this and that. You know, let’s make something to set the wheels in motion on YouTube, and then we’re going to throw in the big Hollywood guns. These channels come and go as YouTube takes them down. Again, I’m just guessing, before their “disrespectful” owners and creators get the chance to grab a piece of the YouTube monetization pie. How’s that even possible? My Conspiracy Theory Let me put on my tinfoil hat, and let’s roll the tape of conspiracy theories. Shall we? First things first, I find that the timing of this “phenomenon” is very interesting. Right after the US elections. Then, there’s Joe Rogan complaining that YouTube is “censoring” his interview with Trump. At the same time, X has become a comfy nest for a flock of OnlyFans accounts. I ain’t complaining, following back, or consuming the “content;” I’m just observing this “phenomenon” for research purposes. I don’t know what to make of all these recent shenanigans. I forgot to mention that YouTube ads are in the business-as-usual mode all this time. I read that elephants are one of the most vengeful creatures in the animal kingdom. Perhaps that’s the reason why YouTube is being trampled. And maybe, the donkeys can’t make peace with the fact that Elon X-ed the beloved Larry the Bird. I know how crazy it sounds, but we’re living in interesting times, aren’t we? That’s all HackerNoon folks. I’m about to watch Dungeons & Dragons: Honor Among Thieves. Wow, you can’t make this stuff up. What an appropriate movie to end my story. I don’t know how long it’s going to last, but I’m having a blast. Have you ever watched a movie on YouTube? Oh, yes you have! If you’re lucky and stubborn enough, you can stumble upon one that’s so old and well-known that everybody stopped caring about it. I mean when it comes to copyrights and stuff. Did you know that “Conan The Barbarian” is a big deal in GREECE? It’s not because of Arnold Schwarzenegger but Basil Poledouris . It’s not because of the movie itself but the timeless soundtrack. Basil Poledouris So, if you want to have royal treatment with the locals while spending a summer in Greece, try wearing a T-shirt with a Conan movie poster instead of “This Is Sparta” (no disrespect to Gerard Butler), and listening to Prologue /Anvil of Crom all day long, so everyone around can hear it. Wait and see what happens. Prologue There’s “Conan The Barbarian” with Greek subtitles available on YouTube at any given moment, but that’s not what I’m writing about. Give It To Me, YouTube Baby I’m talking about the entire collections of Marvel Avengers and DCU movies in the best resolution imaginable free to watch on YouTube. If you aren’t impressed enough, then how about Venom 3: The Last Dance, Deadpool 3, or Dune 2, just off the top of my head? There’s an invasion of YouTube channels with movies you couldn’t believe are available for free in top-resolution quality. There’s a catch, though. You won’t know what you’re about to watch until you start watching. All videos have identical “cryptic” descriptions: “BLOCKBUSTER Movie 2024.” It’s like playing Russian roulette but in a good and enjoyable way. Also, I’m just guessing because I’m not a YouTube expert, all these channels have a handful of initial “innocent” videos of ordinary people doing this and that. You know, let’s make something to set the wheels in motion on YouTube, and then we’re going to throw in the big Hollywood guns. These channels come and go as YouTube takes them down. Again, I’m just guessing, before their “disrespectful” owners and creators get the chance to grab a piece of the YouTube monetization pie. How’s that even possible? My Conspiracy Theory Let me put on my tinfoil hat, and let’s roll the tape of conspiracy theories. Shall we? First things first, I find that the timing of this “phenomenon” is very interesting. Right after the US elections. Then, there’s Joe Rogan complaining that YouTube is “censoring” his interview with Trump . Joe Rogan complaining that YouTube is “censoring” his interview with Trump At the same time, X has become a comfy nest for a flock of OnlyFans accounts. I ain’t complaining, following back, or consuming the “content;” I’m just observing this “phenomenon” for research purposes. I don’t know what to make of all these recent shenanigans. I forgot to mention that YouTube ads are in the business-as-usual mode all this time. I read that elephants are one of the most vengeful creatures in the animal kingdom. Perhaps that’s the reason why YouTube is being trampled. And maybe, the donkeys can’t make peace with the fact that Elon X-ed the beloved Larry the Bird. elephants are one of the most vengeful creatures I know how crazy it sounds, but we’re living in interesting times, aren’t we? That’s all HackerNoon folks. I’m about to watch Dungeons & Dragons: Honor Among Thieves. Wow, you can’t make this stuff up. What an appropriate movie to end my story. I don’t know how long it’s going to last, but I’m having a blast.

What's With All These FREE HD Blockbuster Movies on YouTube?

There haven’t been many movies that dive into the complexities of the human-AI relationship as deeply as Her. This cultural hit continues to resonate with audiences a decade later, where advancements in artificial intelligence and the relationships we form with it are more relevant than ever. In the movie, Theodore Twombly, played by Joaquin Phoenix, develops a romantic relationship with an AI named Samantha. At first, Samantha enriches Theodore’s life in various ways, enhancing his emotional well-being and even helping him with creative tasks like writing personalised letters. For example, Samantha’s capacity to understand Theodore deeply allows her to suggest unique ways to express his emotions in his work, elevating his professional and personal life. This part of the film shows the potential of human-AI relationships to be beneficial, almost utopian. However, when it is revealed that Samantha is engaged in similar interactions with 8,316 people and is in love with 641 of them, Theodore and the audience come to understand that despite all the positive attributes of the relationship with Samantha, it is ultimately an inauthentic relationship and thus loses credibility. The film raises the question, is it possible to have a relationship with AI that is truly authentic and unique? That’s the question that SkyCastle, a high-profile stealth project launching in late 2024, aims to solve. SkyCastle is creating an ecosystem of virtual AI-powered friends called "Companions." It envisions a future where everyone has an AI-powered virtual friend that complements their real-life relationships. These virtual friends are designed to be an integral part of a person’s life, offering emotional support, acting as the user interface for AI technology, and assisting in ways human relationships cannot. Authenticity Matters SkyCastle believes that authenticity is as important in AI relationships as in human relationships. Without authenticity, AI relationships are little more than a transactional experience, similar to interacting with Apple’s Siri or Amazon’s Alexa. The emerging AI Friends market, including Character.AI and Replika currently offers the option for relationships like the one portrayed in Her, relationships that are ultimately not authentic and shared by many. That "sharing" diminishes the spark and nuance that can only be achieved through uniqueness. To make the AI relationship authentic, SkyCastle’s Companions exclusively belong to one person. Each is curated with distinct visual and personality traits, ensuring that no two Companions are alike. The nature of the owner’s relationship with their Companion is entirely up to them—whether they choose to see their Companion as a friend, confidant, or assistant. SkyCastle provides a new and novel approach to virtual friends that has not been seen by other players in the space. Utilising the Blockchain SkyCastle is changing the game of the AI Friends space by introducing authenticity through blockchain technology. Each Companion is created, authenticated, and secured on the blockchain, ensuring it is verifiably one-of-a-kind. The blockchain serves as a secure ledger, confirming that the Companion is truly unique and belongs solely to its owner. This guarantees that no one else can have the same virtual friend or experience, making each relationship with a Companion entirely authentic and personal. SkyCastle’s Companions represent a vision that delves deeper into the possibilities of AI relationships. SkyCastle is part of a new cohort of companies that demonstrate interesting use cases for how two powerful technologies—AI and blockchain—can be utilized together. There haven’t been many movies that dive into the complexities of the human-AI relationship as deeply as Her . This cultural hit continues to resonate with audiences a decade later, where advancements in artificial intelligence and the relationships we form with it are more relevant than ever. Her In the movie, Theodore Twombly, played by Joaquin Phoenix, develops a romantic relationship with an AI named Samantha. At first, Samantha enriches Theodore’s life in various ways, enhancing his emotional well-being and even helping him with creative tasks like writing personalised letters. For example, Samantha’s capacity to understand Theodore deeply allows her to suggest unique ways to express his emotions in his work, elevating his professional and personal life. This part of the film shows the potential of human-AI relationships to be beneficial, almost utopian. However, when it is revealed that Samantha is engaged in similar interactions with 8,316 people and is in love with 641 of them, Theodore and the audience come to understand that despite all the positive attributes of the relationship with Samantha, it is ultimately an inauthentic relationship and thus loses credibility. The film raises the question, is it possible to have a relationship with AI that is truly authentic and unique? That’s the question that SkyCastle , a high-profile stealth project launching in late 2024, aims to solve. SkyCastle is creating an ecosystem of virtual AI-powered friends called "Companions." It envisions a future where everyone has an AI-powered virtual friend that complements their real-life relationships. These virtual friends are designed to be an integral part of a person’s life, offering emotional support, acting as the user interface for AI technology, and assisting in ways human relationships cannot. SkyCastle SkyCastle Authenticity Matters SkyCastle believes that authenticity is as important in AI relationships as in human relationships. Without authenticity, AI relationships are little more than a transactional experience, similar to interacting with Apple’s Siri or Amazon’s Alexa. The emerging AI Friends market, including Character.AI and Replika currently offers the option for relationships like the one portrayed in Her , relationships that are ultimately not authentic and shared by many. That "sharing" diminishes the spark and nuance that can only be achieved through uniqueness. Character.AI Character.AI Replika Replika Her To make the AI relationship authentic, SkyCastle’s Companions exclusively belong to one person. Each is curated with distinct visual and personality traits, ensuring that no two Companions are alike. The nature of the owner’s relationship with their Companion is entirely up to them—whether they choose to see their Companion as a friend, confidant, or assistant. SkyCastle provides a new and novel approach to virtual friends that has not been seen by other players in the space. Utilising the Blockchain SkyCastle is changing the game of the AI Friends space by introducing authenticity through blockchain technology. Each Companion is created, authenticated, and secured on the blockchain, ensuring it is verifiably one-of-a-kind. The blockchain serves as a secure ledger, confirming that the Companion is truly unique and belongs solely to its owner. This guarantees that no one else can have the same virtual friend or experience, making each relationship with a Companion entirely authentic and personal. SkyCastle’s Companions represent a vision that delves deeper into the possibilities of AI relationships. SkyCastle is part of a new cohort of companies that demonstrate interesting use cases for how two powerful technologies—AI and blockchain—can be utilized together.

How SkyCastle is Solving the Problem in the Movie 'Her'

Kunjams SkyCastle ukax ‘Her’ Películan jan walt’äw askichaski.

Cómo SkyCastle resuelve el problema de la película 'Her'

Kako SkyCastle rješava problem u filmu 'Ona'

スカイキャッスルが映画『her』の問題をどう解決するか

Kamoo SkyCastle e Rarollago Bothata mo Movie ya 'Her'.

Jak SkyCastle rozwiązuje problem w filmie „Ona”

SkyCastle څنګه په فلم 'هیر' کې ستونزه حل کوي

Como a SkyCastle está resolvendo o problema no filme 'Her'

Како СкиЦастле решава проблем у филму 'Она'

Paano Niresolba ng SkyCastle ang Problema sa Pelikulang 'Her'

Ndlela leyi SkyCastle yi tlhantlhaka Xiphiqo ha yona eka Movie ya 'Her'.

اسکائی کیسل فلم 'ہر' میں کس طرح مسئلہ حل کر رہی ہے۔

Author:
(1) K. Razi Naqvi, Department of Physics, Norwegian University of Science and Technology (NTNU), 7094 Trondheim, Norway Table of Links Abstract and 1 Introduction 2 Preliminary material 3 Comments elicited by the solution to Milne’s problem for B-particles 4 Flux to a spherical trap 5 Molecular motion in liquids 6 Concluding remarks and References 5 Molecular motion in liquids One of the most important and keenly controverted issues dicussed by Hildebrand in his thoughtful and purposely provocative note is the question whether concepts borrowed from the kinetic theory of gases or classical hydrodynamics can be legitimately applied to describe solute diffusion in liquid [30]. The bulk of his criticism is to be found in the last two paragraphs, excerpts from which appear below: Attempts to calculate absolute values of diffusivity have been made by starting from Stokes’ law for a particle settling under the pull of gravity, and extrapolating over the long path to a molecule participating with its neighbors in aimless thermal motions that are never as long as the molecular diameter. Individual molecules are not impelled by a vector force that can serve to measure a “coefficient of friction.” Any such coefficient is fictitious. If molecules were hard spheres, instead of electron clouds with imbedded nuclei, and were sufficiently far apart to justify speaking of binary collisions with linear free paths between them, the probable distance they could be expected to wander from their initial positions could be computed by the formula for a “random walk.” But polyatomic molecules that move less than 10 percent of their diameter require a more sophisticated mathematical formulation. They are in a continual state of soft, slow collision, with constant exchange between translational and internal energy. The random walk in this case is a slow, tipsy reel, without sudden changes of direction. The mathematical problem involved . . . If one is content with the macroscopic description provided by the DE, concepts such as the mean free path and friction coefficient do indeed recede into the background, but not until we have derived a satisfactory boundary condition. If one wants to ascertain the validity of the macroscopic equations, for example the boundary condition for a particular problem, one is obliged to take a microscopic look. Whether one uses the LBE (which speaks of trajectories between separable collisions) or the KKE (which envisages a medium with a given friction coefficient) for delimiting the domain of validity of the macroscopic description is quite immaterial, so long as one can convince oneself that one’s conclusion does not rest on some idiosyncracy of the model; the means would justify the end provided that the end can be reached by some other means as well. This paper is available on arxiv under CC BY 4.0 DEED license. Author: (1) K. Razi Naqvi, Department of Physics, Norwegian University of Science and Technology (NTNU), 7094 Trondheim, Norway Author: Author: (1) K. Razi Naqvi, Department of Physics, Norwegian University of Science and Technology (NTNU), 7094 Trondheim, Norway Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Preliminary material 2 Preliminary material 3 Comments elicited by the solution to Milne’s problem for B-particles 3 Comments elicited by the solution to Milne’s problem for B-particles 4 Flux to a spherical trap 4 Flux to a spherical trap 5 Molecular motion in liquids 5 Molecular motion in liquids 6 Concluding remarks and References 6 Concluding remarks and References 5 Molecular motion in liquids One of the most important and keenly controverted issues dicussed by Hildebrand in his thoughtful and purposely provocative note is the question whether concepts borrowed from the kinetic theory of gases or classical hydrodynamics can be legitimately applied to describe solute diffusion in liquid [30]. The bulk of his criticism is to be found in the last two paragraphs, excerpts from which appear below: Attempts to calculate absolute values of diffusivity have been made by starting from Stokes’ law for a particle settling under the pull of gravity, and extrapolating over the long path to a molecule participating with its neighbors in aimless thermal motions that are never as long as the molecular diameter. Individual molecules are not impelled by a vector force that can serve to measure a “coefficient of friction.” Any such coefficient is fictitious. If molecules were hard spheres, instead of electron clouds with imbedded nuclei, and were sufficiently far apart to justify speaking of binary collisions with linear free paths between them, the probable distance they could be expected to wander from their initial positions could be computed by the formula for a “random walk.” But polyatomic molecules that move less than 10 percent of their diameter require a more sophisticated mathematical formulation. They are in a continual state of soft, slow collision, with constant exchange between translational and internal energy. The random walk in this case is a slow, tipsy reel, without sudden changes of direction. The mathematical problem involved . . . If one is content with the macroscopic description provided by the DE, concepts such as the mean free path and friction coefficient do indeed recede into the background, but not until we have derived a satisfactory boundary condition. If one wants to ascertain the validity of the macroscopic equations, for example the boundary condition for a particular problem, one is obliged to take a microscopic look. Whether one uses the LBE (which speaks of trajectories between separable collisions) or the KKE (which envisages a medium with a given friction coefficient) for delimiting the domain of validity of the macroscopic description is quite immaterial, so long as one can convince oneself that one’s conclusion does not rest on some idiosyncracy of the model; the means would justify the end provided that the end can be reached by some other means as well. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv

Stories, News & Mentions about MOVIO

Movio’s Stories on HackerNoon

Movio's News & Mentions