Amit H. Bermano

Rinon Gal, Yuval Alaluf, Yuval Atzmon et al. · nvidia

Text-to-image models offer unprecedented freedom to guide creation through natural language. Yet, it is unclear how such freedom can be exercised to generate images of specific unique concepts, modify their appearance, or compose them in new roles and novel scenes. In other words, we ask: how can we use language-guided models to turn our cat into a painting, or imagine a new product based on our favorite toy? Here we present a simple approach that allows such creative freedom. Using only 3-5 images of a user-provided concept, like an object or a style, we learn to represent it through new "words" in the embedding space of a frozen text-to-image model. These "words" can be composed into natural language sentences, guiding personalized creation in an intuitive way. Notably, we find evidence that a single word embedding is sufficient for capturing unique and varied concepts. We compare our approach to a wide range of baselines, and demonstrate that it can more faithfully portray the concepts across a range of applications and tasks. Our code, data and new words will be available at: https://textual-inversion.github.io

Amir Barda, Matheus Gadelha, Vladimir G. Kim et al.

We propose a generative technique to edit 3D shapes, represented as meshes, NeRFs, or Gaussian Splats, in approximately 3 seconds, without the need for running an SDS type of optimization. Our key insight is to cast 3D editing as a multiview image inpainting problem, as this representation is generic and can be mapped back to any 3D representation using the bank of available Large Reconstruction Models. We explore different fine-tuning strategies to obtain both multiview generation and inpainting capabilities within the same diffusion model. In particular, the design of the inpainting mask is an important factor of training an inpainting model, and we propose several masking strategies to mimic the types of edits a user would perform on a 3D shape. Our approach takes 3D generative editing from hours to seconds and produces higher-quality results compared to previous works.

Guy Tevet, Brian Gordon, Amir Hertz et al.

We introduce MotionCLIP, a 3D human motion auto-encoder featuring a latent embedding that is disentangled, well behaved, and supports highly semantic textual descriptions. MotionCLIP gains its unique power by aligning its latent space with that of the Contrastive Language-Image Pre-training (CLIP) model. Aligning the human motion manifold to CLIP space implicitly infuses the extremely rich semantic knowledge of CLIP into the manifold. In particular, it helps continuity by placing semantically similar motions close to one another, and disentanglement, which is inherited from the CLIP-space structure. MotionCLIP comprises a transformer-based motion auto-encoder, trained to reconstruct motion while being aligned to its text label's position in CLIP-space. We further leverage CLIP's unique visual understanding and inject an even stronger signal through aligning motion to rendered frames in a self-supervised manner. We show that although CLIP has never seen the motion domain, MotionCLIP offers unprecedented text-to-motion abilities, allowing out-of-domain actions, disentangled editing, and abstract language specification. For example, the text prompt "couch" is decoded into a sitting down motion, due to lingual similarity, and the prompt "Spiderman" results in a web-swinging-like solution that is far from seen during training. In addition, we show how the introduced latent space can be leveraged for motion interpolation, editing and recognition.

Rinon Gal, Moab Arar, Yuval Atzmon et al.

Text-to-image personalization aims to teach a pre-trained diffusion model to reason about novel, user provided concepts, embedding them into new scenes guided by natural language prompts. However, current personalization approaches struggle with lengthy training times, high storage requirements or loss of identity. To overcome these limitations, we propose an encoder-based domain-tuning approach. Our key insight is that by underfitting on a large set of concepts from a given domain, we can improve generalization and create a model that is more amenable to quickly adding novel concepts from the same domain. Specifically, we employ two components: First, an encoder that takes as an input a single image of a target concept from a given domain, e.g. a specific face, and learns to map it into a word-embedding representing the concept. Second, a set of regularized weight-offsets for the text-to-image model that learn how to effectively ingest additional concepts. Together, these components are used to guide the learning of unseen concepts, allowing us to personalize a model using only a single image and as few as 5 training steps - accelerating personalization from dozens of minutes to seconds, while preserving quality.

Ryan Po, Wang Yifan, Vladislav Golyanik et al.

The field of visual computing is rapidly advancing due to the emergence of generative artificial intelligence (AI), which unlocks unprecedented capabilities for the generation, editing, and reconstruction of images, videos, and 3D scenes. In these domains, diffusion models are the generative AI architecture of choice. Within the last year alone, the literature on diffusion-based tools and applications has seen exponential growth and relevant papers are published across the computer graphics, computer vision, and AI communities with new works appearing daily on arXiv. This rapid growth of the field makes it difficult to keep up with all recent developments. The goal of this state-of-the-art report (STAR) is to introduce the basic mathematical concepts of diffusion models, implementation details and design choices of the popular Stable Diffusion model, as well as overview important aspects of these generative AI tools, including personalization, conditioning, inversion, among others. Moreover, we give a comprehensive overview of the rapidly growing literature on diffusion-based generation and editing, categorized by the type of generated medium, including 2D images, videos, 3D objects, locomotion, and 4D scenes. Finally, we discuss available datasets, metrics, open challenges, and social implications. This STAR provides an intuitive starting point to explore this exciting topic for researchers, artists, and practitioners alike.

Moab Arar, Rinon Gal, Yuval Atzmon et al.

Text-to-image (T2I) personalization allows users to guide the creative image generation process by combining their own visual concepts in natural language prompts. Recently, encoder-based techniques have emerged as a new effective approach for T2I personalization, reducing the need for multiple images and long training times. However, most existing encoders are limited to a single-class domain, which hinders their ability to handle diverse concepts. In this work, we propose a domain-agnostic method that does not require any specialized dataset or prior information about the personalized concepts. We introduce a novel contrastive-based regularization technique to maintain high fidelity to the target concept characteristics while keeping the predicted embeddings close to editable regions of the latent space, by pushing the predicted tokens toward their nearest existing CLIP tokens. Our experimental results demonstrate the effectiveness of our approach and show how the learned tokens are more semantic than tokens predicted by unregularized models. This leads to a better representation that achieves state-of-the-art performance while being more flexible than previous methods.

Guy Tevet, Sigal Raab, Brian Gordon et al.

Natural and expressive human motion generation is the holy grail of computer animation. It is a challenging task, due to the diversity of possible motion, human perceptual sensitivity to it, and the difficulty of accurately describing it. Therefore, current generative solutions are either low-quality or limited in expressiveness. Diffusion models, which have already shown remarkable generative capabilities in other domains, are promising candidates for human motion due to their many-to-many nature, but they tend to be resource hungry and hard to control. In this paper, we introduce Motion Diffusion Model (MDM), a carefully adapted classifier-free diffusion-based generative model for the human motion domain. MDM is transformer-based, combining insights from motion generation literature. A notable design-choice is the prediction of the sample, rather than the noise, in each diffusion step. This facilitates the use of established geometric losses on the locations and velocities of the motion, such as the foot contact loss. As we demonstrate, MDM is a generic approach, enabling different modes of conditioning, and different generation tasks. We show that our model is trained with lightweight resources and yet achieves state-of-the-art results on leading benchmarks for text-to-motion and action-to-motion. https://guytevet.github.io/mdm-page/ .

Roy Kapon, Guy Tevet, Daniel Cohen-Or et al.

We introduce Multi-view Ancestral Sampling (MAS), a method for 3D motion generation, using 2D diffusion models that were trained on motions obtained from in-the-wild videos. As such, MAS opens opportunities to exciting and diverse fields of motion previously under-explored as 3D data is scarce and hard to collect. MAS works by simultaneously denoising multiple 2D motion sequences representing different views of the same 3D motion. It ensures consistency across all views at each diffusion step by combining the individual generations into a unified 3D sequence, and projecting it back to the original views. We demonstrate MAS on 2D pose data acquired from videos depicting professional basketball maneuvers, rhythmic gymnastic performances featuring a ball apparatus, and horse races. In each of these domains, 3D motion capture is arduous, and yet, MAS generates diverse and realistic 3D sequences. Unlike the Score Distillation approach, which optimizes each sample by repeatedly applying small fixes, our method uses a sampling process that was constructed for the diffusion framework. As we demonstrate, MAS avoids common issues such as out-of-domain sampling and mode-collapse. https://guytevet.github.io/mas-page/

Yonatan Shafir, Guy Tevet, Roy Kapon et al.

Recent work has demonstrated the significant potential of denoising diffusion models for generating human motion, including text-to-motion capabilities. However, these methods are restricted by the paucity of annotated motion data, a focus on single-person motions, and a lack of detailed control. In this paper, we introduce three forms of composition based on diffusion priors: sequential, parallel, and model composition. Using sequential composition, we tackle the challenge of long sequence generation. We introduce DoubleTake, an inference-time method with which we generate long animations consisting of sequences of prompted intervals and their transitions, using a prior trained only for short clips. Using parallel composition, we show promising steps toward two-person generation. Beginning with two fixed priors as well as a few two-person training examples, we learn a slim communication block, ComMDM, to coordinate interaction between the two resulting motions. Lastly, using model composition, we first train individual priors to complete motions that realize a prescribed motion for a given joint. We then introduce DiffusionBlending, an interpolation mechanism to effectively blend several such models to enable flexible and efficient fine-grained joint and trajectory-level control and editing. We evaluate the composition methods using an off-the-shelf motion diffusion model, and further compare the results to dedicated models trained for these specific tasks.

Sigal Raab, Inbal Leibovitch, Guy Tevet et al.

Synthesizing realistic animations of humans, animals, and even imaginary creatures, has long been a goal for artists and computer graphics professionals. Compared to the imaging domain, which is rich with large available datasets, the number of data instances for the motion domain is limited, particularly for the animation of animals and exotic creatures (e.g., dragons), which have unique skeletons and motion patterns. In this work, we present a Single Motion Diffusion Model, dubbed SinMDM, a model designed to learn the internal motifs of a single motion sequence with arbitrary topology and synthesize motions of arbitrary length that are faithful to them. We harness the power of diffusion models and present a denoising network explicitly designed for the task of learning from a single input motion. SinMDM is designed to be a lightweight architecture, which avoids overfitting by using a shallow network with local attention layers that narrow the receptive field and encourage motion diversity. SinMDM can be applied in various contexts, including spatial and temporal in-betweening, motion expansion, style transfer, and crowd animation. Our results show that SinMDM outperforms existing methods both in quality and time-space efficiency. Moreover, while current approaches require additional training for different applications, our work facilitates these applications at inference time. Our code and trained models are available at https://sinmdm.github.io/SinMDM-page.

Ben Maman, Amit H. Bermano

Multi-instrument Automatic Music Transcription (AMT), or the decoding of a musical recording into semantic musical content, is one of the holy grails of Music Information Retrieval. Current AMT approaches are restricted to piano and (some) guitar recordings, due to difficult data collection. In order to overcome data collection barriers, previous AMT approaches attempt to employ musical scores in the form of a digitized version of the same song or piece. The scores are typically aligned using audio features and strenuous human intervention to generate training labels. We introduce NoteEM, a method for simultaneously training a transcriber and aligning the scores to their corresponding performances, in a fully-automated process. Using this unaligned supervision scheme, complemented by pseudo-labels and pitch-shift augmentation, our method enables training on in-the-wild recordings with unprecedented accuracy and instrumental variety. Using only synthetic data and unaligned supervision, we report SOTA note-level accuracy of the MAPS dataset, and large favorable margins on cross-dataset evaluations. We also demonstrate robustness and ease of use; we report comparable results when training on a small, easily obtainable, self-collected dataset, and we propose alternative labeling to the MusicNet dataset, which we show to be more accurate. Our project page is available at https://benadar293.github.io

Rinon Gal, Yael Vinker, Yuval Alaluf et al.

A sketch is one of the most intuitive and versatile tools humans use to convey their ideas visually. An animated sketch opens another dimension to the expression of ideas and is widely used by designers for a variety of purposes. Animating sketches is a laborious process, requiring extensive experience and professional design skills. In this work, we present a method that automatically adds motion to a single-subject sketch (hence, "breathing life into it"), merely by providing a text prompt indicating the desired motion. The output is a short animation provided in vector representation, which can be easily edited. Our method does not require extensive training, but instead leverages the motion prior of a large pretrained text-to-video diffusion model using a score-distillation loss to guide the placement of strokes. To promote natural and smooth motion and to better preserve the sketch's appearance, we model the learned motion through two components. The first governs small local deformations and the second controls global affine transformations. Surprisingly, we find that even models that struggle to generate sketch videos on their own can still serve as a useful backbone for animating abstract representations.

Haim Sawdayee, Amir Vaxman, Amit H. Bermano

Reconstructing 3D shapes from planar cross-sections is a challenge inspired by downstream applications like medical imaging and geographic informatics. The input is an in/out indicator function fully defined on a sparse collection of planes in space, and the output is an interpolation of the indicator function to the entire volume. Previous works addressing this sparse and ill-posed problem either produce low quality results, or rely on additional priors such as target topology, appearance information, or input normal directions. In this paper, we present OReX, a method for 3D shape reconstruction from slices alone, featuring a Neural Field as the interpolation prior. A modest neural network is trained on the input planes to return an inside/outside estimate for a given 3D coordinate, yielding a powerful prior that induces smoothness and self-similarities. The main challenge for this approach is high-frequency details, as the neural prior is overly smoothing. To alleviate this, we offer an iterative estimation architecture and a hierarchical input sampling scheme that encourage coarse-to-fine training, allowing the training process to focus on high frequencies at later stages. In addition, we identify and analyze a ripple-like effect stemming from the mesh extraction step. We mitigate it by regularizing the spatial gradients of the indicator function around input in/out boundaries during network training, tackling the problem at the root. Through extensive qualitative and quantitative experimentation, we demonstrate our method is robust, accurate, and scales well with the size of the input. We report state-of-the-art results compared to previous approaches and recent potential solutions, and demonstrate the benefit of our individual contributions through analysis and ablation studies.

Yotam Erel, Daisuke Iwai, Amit H. Bermano

We introduce a high resolution spatially adaptive light source, or a projector, into a neural reflectance field that allows to both calibrate the projector and photo realistic light editing. The projected texture is fully differentiable with respect to all scene parameters, and can be optimized to yield a desired appearance suitable for applications in augmented reality and projection mapping. Our neural field consists of three neural networks, estimating geometry, material, and transmittance. Using an analytical BRDF model and carefully selected projection patterns, our acquisition process is simple and intuitive, featuring a fixed uncalibrated projected and a handheld camera with a co-located light source. As we demonstrate, the virtual projector incorporated into the pipeline improves scene understanding and enables various projection mapping applications, alleviating the need for time consuming calibration steps performed in a traditional setting per view or projector location. In addition to enabling novel viewpoint synthesis, we demonstrate state-of-the-art performance projector compensation for novel viewpoints, improvement over the baselines in material and scene reconstruction, and three simply implemented scenarios where projection image optimization is performed, including the use of a 2D generative model to consistently dictate scene appearance from multiple viewpoints. We believe that neural projection mapping opens up the door to novel and exciting downstream tasks, through the joint optimization of the scene and projection images.

Ben Maman, Johannes Zeitler, Meinard Müller et al.

Generating multi-instrument music from symbolic music representations is an important task in Music Information Retrieval (MIR). A central but still largely unsolved problem in this context is musically and acoustically informed control in the generation process. As the main contribution of this work, we propose enhancing control of multi-instrument synthesis by conditioning a generative model on a specific performance and recording environment, thus allowing for better guidance of timbre and style. Building on state-of-the-art diffusion-based music generative models, we introduce performance conditioning - a simple tool indicating the generative model to synthesize music with style and timbre of specific instruments taken from specific performances. Our prototype is evaluated using uncurated performances with diverse instrumentation and achieves state-of-the-art FAD realism scores while allowing novel timbre and style control. Our project page, including samples and demonstrations, is available at benadar293.github.io/midipm

Shahar Sarfaty, Adi Haviv, Uri Hacohen et al.

The rapid proliferation of generative components, such as LoRAs, has created a vast but unstructured ecosystem. Existing discovery methods depend on unreliable user descriptions or biased popularity metrics, hindering usability. We present CARLoS, a large-scale framework for characterizing LoRAs without requiring additional metadata. Analyzing over 650 LoRAs, we employ them in image generation over a variety of prompts and seeds, as a credible way to assess their behavior. Using CLIP embeddings and their difference to a base-model generation, we concisely define a three-part representation: Directions, defining semantic shift; Strength, quantifying the significance of the effect; and Consistency, quantifying how stable the effect is. Using these representations, we develop an efficient retrieval framework that semantically matches textual queries to relevant LoRAs while filtering overly strong or unstable ones, outperforming textual baselines in automated and human evaluations. While retrieval is our primary focus, the same representation also supports analyses linking Strength and Consistency to legal notions of substantiality and volition, key considerations in copyright, positioning CARLoS as a practical system with broader relevance for LoRA analysis.

Ofir Bar Tal, Adi Haviv, Amit H. Bermano

Evasion Attacks (EA) are used to test the robustness of trained neural networks by distorting input data to misguide the model into incorrect classifications. Creating these attacks is a challenging task, especially with the ever-increasing complexity of models and datasets. In this work, we introduce a self-supervised, computationally economical method for generating adversarial examples, designed for the unseen black-box setting. Adapting techniques from representation learning, our method generates on-manifold EAs that are encouraged to resemble the data distribution. These attacks are comparable in effectiveness compared to the state-of-the-art when attacking the model trained on, but are significantly more effective when attacking unseen models, as the attacks are more related to the data rather than the model itself. Our experiments consistently demonstrate the method is effective across various models, unseen data categories, and even defended models, suggesting a significant role for on-manifold EAs when targeting unseen models.

Yunzhe Liu, Rinon Gal, Amit H. Bermano et al.

Generative Adversarial Networks (GANs) are susceptible to bias, learned from either the unbalanced data, or through mode collapse. The networks focus on the core of the data distribution, leaving the tails - or the edges of the distribution - behind. We argue that this bias is responsible not only for fairness concerns, but that it plays a key role in the collapse of latent-traversal editing methods when deviating away from the distribution's core. Building on this observation, we outline a method for mitigating generative bias through a self-conditioning process, where distances in the latent-space of a pre-trained generator are used to provide initial labels for the data. By fine-tuning the generator on a re-sampled distribution drawn from these self-labeled data, we force the generator to better contend with rare semantic attributes and enable more realistic generation of these properties. We compare our models to a wide range of latent editing methods, and show that by alleviating the bias they achieve finer semantic control and better identity preservation through a wider range of transformations. Our code and models will be available at https://github.com/yzliu567/sc-gan

Moab Arar, Ariel Shamir, Amit H. Bermano

Vision Transformers (ViT) serve as powerful vision models. Unlike convolutional neural networks, which dominated vision research in previous years, vision transformers enjoy the ability to capture long-range dependencies in the data. Nonetheless, an integral part of any transformer architecture, the self-attention mechanism, suffers from high latency and inefficient memory utilization, making it less suitable for high-resolution input images. To alleviate these shortcomings, hierarchical vision models locally employ self-attention on non-interleaving windows. This relaxation reduces the complexity to be linear in the input size; however, it limits the cross-window interaction, hurting the model performance. In this paper, we propose a new shift-invariant local attention layer, called query and attend (QnA), that aggregates the input locally in an overlapping manner, much like convolutions. The key idea behind QnA is to introduce learned queries, which allow fast and efficient implementation. We verify the effectiveness of our layer by incorporating it into a hierarchical vision transformer model. We show improvements in speed and memory complexity while achieving comparable accuracy with state-of-the-art models. Finally, our layer scales especially well with window size, requiring up-to x10 less memory while being up-to x5 faster than existing methods. The code is publicly available at \url{https://github.com/moabarar/qna}.

Ron Mokady, Amir Hertz, Amit H. Bermano

Image captioning is a fundamental task in vision-language understanding, where the model predicts a textual informative caption to a given input image. In this paper, we present a simple approach to address this task. We use CLIP encoding as a prefix to the caption, by employing a simple mapping network, and then fine-tunes a language model to generate the image captions. The recently proposed CLIP model contains rich semantic features which were trained with textual context, making it best for vision-language perception. Our key idea is that together with a pre-trained language model (GPT2), we obtain a wide understanding of both visual and textual data. Hence, our approach only requires rather quick training to produce a competent captioning model. Without additional annotations or pre-training, it efficiently generates meaningful captions for large-scale and diverse datasets. Surprisingly, our method works well even when only the mapping network is trained, while both CLIP and the language model remain frozen, allowing a lighter architecture with less trainable parameters. Through quantitative evaluation, we demonstrate our model achieves comparable results to state-of-the-art methods on the challenging Conceptual Captions and nocaps datasets, while it is simpler, faster, and lighter. Our code is available in https://github.com/rmokady/CLIP_prefix_caption.

Rinon Gal, Or Lichter, Elad Richardson et al.

Recent advancements in diffusion models have introduced fast sampling methods that can effectively produce high-quality images in just one or a few denoising steps. Interestingly, when these are distilled from existing diffusion models, they often maintain alignment with the original model, retaining similar outputs for similar prompts and seeds. These properties present opportunities to leverage fast sampling methods as a shortcut-mechanism, using them to create a preview of denoised outputs through which we can backpropagate image-space losses. In this work, we explore the potential of using such shortcut-mechanisms to guide the personalization of text-to-image models to specific facial identities. We focus on encoder-based personalization approaches, and demonstrate that by tuning them with a lookahead identity loss, we can achieve higher identity fidelity, without sacrificing layout diversity or prompt alignment. We further explore the use of attention sharing mechanisms and consistent data generation for the task of personalization, and find that encoder training can benefit from both.

Inbar Gat, Sigal Raab, Guy Tevet et al.

Generating motion for arbitrary skeletons is a longstanding challenge in computer graphics, remaining largely unexplored due to the scarcity of diverse datasets and the irregular nature of the data. In this work, we introduce AnyTop, a diffusion model that generates motions for diverse characters with distinct motion dynamics, using only their skeletal structure as input. Our work features a transformer-based denoising network, tailored for arbitrary skeleton learning, integrating topology information into the traditional attention mechanism. Additionally, by incorporating textual joint descriptions into the latent feature representation, AnyTop learns semantic correspondences between joints across diverse skeletons. Our evaluation demonstrates that AnyTop generalizes well, even with as few as three training examples per topology, and can produce motions for unseen skeletons as well. Furthermore, our model's latent space is highly informative, enabling downstream tasks such as joint correspondence, temporal segmentation and motion editing. Our webpage, https://anytop2025.github.io/Anytop-page, includes links to videos and code.

Rotem Shalev-Arkushin, Rinon Gal, Amit H. Bermano et al.

Diffusion models enable high-quality and diverse visual content synthesis. However, they struggle to generate rare or unseen concepts. To address this challenge, we explore the usage of Retrieval-Augmented Generation (RAG) with image generation models. We propose ImageRAG, a method that dynamically retrieves relevant images based on a given text prompt, and uses them as context to guide the generation process. Prior approaches that used retrieved images to improve generation, trained models specifically for retrieval-based generation. In contrast, ImageRAG leverages the capabilities of existing image conditioning models, and does not require RAG-specific training. Our approach is highly adaptable and can be applied across different model types, showing significant improvement in generating rare and fine-grained concepts using different base models. Our project page is available at: https://rotem-shalev.github.io/ImageRAG

Haim Sawdayee, Chuan Guo, Guy Tevet et al.

Text-to-motion generative models span a wide range of 3D human actions but struggle with nuanced stylistic attributes such as a "Chicken" style. Due to the scarcity of style-specific data, existing approaches pull the generative prior towards a reference style, which often results in out-of-distribution low quality generations. In this work, we introduce LoRA-MDM, a lightweight framework for motion stylization that generalizes to complex actions while maintaining editability. Our key insight is that adapting the generative prior to include the style, while preserving its overall distribution, is more effective than modifying each individual motion during generation. Building on this idea, LoRA-MDM learns to adapt the prior to include the reference style using only a few samples. The style can then be used in the context of different textual prompts for generation. The low-rank adaptation shifts the motion manifold in a semantically meaningful way, enabling realistic style infusion even for actions not present in the reference samples. Moreover, preserving the distribution structure enables advanced operations such as style blending and motion editing. We compare LoRA-MDM to state-of-the-art stylized motion generation methods and demonstrate a favorable balance between text fidelity and style consistency.

Roey Ron, Guy Tevet, Haim Sawdayee et al.

We present HOIDiNi, a text-driven diffusion framework for synthesizing realistic and plausible human-object interaction (HOI). HOI generation is extremely challenging since it induces strict contact accuracies alongside a diverse motion manifold. While current literature trades off between realism and physical correctness, HOIDiNi optimizes directly in the noise space of a pretrained diffusion model using Diffusion Noise Optimization (DNO), achieving both. This is made feasible thanks to our observation that the problem can be separated into two phases: an object-centric phase, primarily making discrete choices of hand-object contact locations, and a human-centric phase that refines the full-body motion to realize this blueprint. This structured approach allows for precise hand-object contact without compromising motion naturalness. Quantitative, qualitative, and subjective evaluations on the GRAB dataset alone clearly indicate HOIDiNi outperforms prior works and baselines in contact accuracy, physical validity, and overall quality. Our results demonstrate the ability to generate complex, controllable interactions, including grasping, placing, and full-body coordination, driven solely by textual prompts. https://hoidini.github.io.

Jonathan Yaffe, Ben Maman, Meinard Müller et al.

Automatic Music Transcription (AMT) converts audio recordings into symbolic musical representations. Training deep neural networks (DNNs) for AMT typically requires strongly aligned training pairs with precise frame-level annotations. Since creating such datasets is costly and impractical for many musical contexts, weakly aligned approaches using segment-level annotations have gained traction. However, existing methods often rely on Dynamic Time Warping (DTW) or soft alignment loss functions, both of which still require local semantic correspondences, making them error-prone and computationally expensive. In this article, we introduce CountEM, a novel AMT framework that eliminates the need for explicit local alignment by leveraging note event histograms as supervision, enabling lighter computations and greater flexibility. Using an Expectation-Maximization (EM) approach, CountEM iteratively refines predictions based solely on note occurrence counts, significantly reducing annotation efforts while maintaining high transcription accuracy. Experiments on piano, guitar, and multi-instrument datasets demonstrate that CountEM matches or surpasses existing weakly supervised methods, improving AMT's robustness, scalability, and efficiency. Our project page is available at https://yoni-yaffe.github.io/count-the-notes.

Yotam Erel, Rishabh Dabral, Vladislav Golyanik et al.

Light control in generated images is a difficult task, posing specific challenges, spanning over the entire image and frequency spectrum. Most approaches tackle this problem by training on extensive yet domain-specific datasets, limiting the inherent generalization and applicability of the foundational backbones used. Instead, PractiLight is a practical approach, effectively leveraging foundational understanding of recent generative models for the task. Our key insight is that lighting relationships in an image are similar in nature to token interaction in self-attention layers, and hence are best represented there. Based on this and other analyses regarding the importance of early diffusion iterations, PractiLight trains a lightweight LoRA regressor to produce the direct irradiance map for a given image, using a small set of training images. We then employ this regressor to incorporate the desired lighting into the generation process of another image using Classifier Guidance. This careful design generalizes well to diverse conditions and image domains. We demonstrate state-of-the-art performance in terms of quality and control with proven parameter and data efficiency compared to leading works over a wide variety of scenes types. We hope this work affirms that image lighting can feasibly be controlled by tapping into foundational knowledge, enabling practical and general relighting.

Yotam Erel, Olaf Dünkel, Rishabh Dabral et al.

We introduce a new interpretation of the attention matrix as a discrete-time Markov chain. Our interpretation sheds light on common operations involving attention scores such as selection, summation, and averaging in a unified framework. It further extends them by considering indirect attention, propagated through the Markov chain, as opposed to previous studies that only model immediate effects. Our key observation is that tokens linked to semantically similar regions form metastable states, i.e., regions where attention tends to concentrate, while noisy attention scores dissipate. Metastable states and their prevalence can be easily computed through simple matrix multiplication and eigenanalysis, respectively. Using these lightweight tools, we demonstrate state-of-the-art zero-shot segmentation. Lastly, we define TokenRank -- the steady state vector of the Markov chain, which measures global token importance. We show that TokenRank enhances unconditional image generation, improving both quality (IS) and diversity (FID), and can also be incorporated into existing segmentation techniques to improve their performance over existing benchmarks. We believe our framework offers a fresh view of how tokens are being attended in modern visual transformers.

Nadav Orzech, Yotam Nitzan, Ulysse Mizrahi et al.

Virtual Try-On (VTON) is a highly active line of research, with increasing demand. It aims to replace a piece of garment in an image with one from another, while preserving person and garment characteristics as well as image fidelity. Current literature takes a supervised approach for the task, impairing generalization and imposing heavy computation. In this paper, we present a novel zero-shot training-free method for inpainting a clothing garment by reference. Our approach employs the prior of a diffusion model with no additional training, fully leveraging its native generalization capabilities. The method employs extended attention to transfer image information from reference to target images, overcoming two significant challenges. We first initially warp the reference garment over the target human using deep features, alleviating "texture sticking". We then leverage the extended attention mechanism with careful masking, eliminating leakage of reference background and unwanted influence. Through a user study, qualitative, and quantitative comparison to state-of-the-art approaches, we demonstrate superior image quality and garment preservation compared unseen clothing pieces or human figures.

Rotem Shalev-Arkushin, Aharon Azulay, Tavi Halperin et al.

Diffusion-based generative models have recently shown remarkable image and video editing capabilities. However, local video editing, particularly removal of small attributes like glasses, remains a challenge. Existing methods either alter the videos excessively, generate unrealistic artifacts, or fail to perform the requested edit consistently throughout the video. In this work, we focus on consistent and identity-preserving removal of glasses in videos, using it as a case study for consistent local attribute removal in videos. Due to the lack of paired data, we adopt a weakly supervised approach and generate synthetic imperfect data, using an adjusted pretrained diffusion model. We show that despite data imperfection, by learning from our generated data and leveraging the prior of pretrained diffusion models, our model is able to perform the desired edit consistently while preserving the original video content. Furthermore, we exemplify the generalization ability of our method to other local video editing tasks by applying it successfully to facial sticker-removal. Our approach demonstrates significant improvement over existing methods, showcasing the potential of leveraging synthetic data and strong video priors for local video editing tasks.

Sigal Raab, Inbar Gat, Nathan Sala et al.

Given the remarkable results of motion synthesis with diffusion models, a natural question arises: how can we effectively leverage these models for motion editing? Existing diffusion-based motion editing methods overlook the profound potential of the prior embedded within the weights of pre-trained models, which enables manipulating the latent feature space; hence, they primarily center on handling the motion space. In this work, we explore the attention mechanism of pre-trained motion diffusion models. We uncover the roles and interactions of attention elements in capturing and representing intricate human motion patterns, and carefully integrate these elements to transfer a leader motion to a follower one while maintaining the nuanced characteristics of the follower, resulting in zero-shot motion transfer. Editing features associated with selected motions allows us to confront a challenge observed in prior motion diffusion approaches, which use general directives (e.g., text, music) for editing, ultimately failing to convey subtle nuances effectively. Our work is inspired by how a monkey closely imitates what it sees while maintaining its unique motion patterns; hence we call it Monkey See, Monkey Do, and dub it MoMo. Employing our technique enables accomplishing tasks such as synthesizing out-of-distribution motions, style transfer, and spatial editing. Furthermore, diffusion inversion is seldom employed for motions; as a result, editing efforts focus on generated motions, limiting the editability of real ones. MoMo harnesses motion inversion, extending its application to both real and generated motions. Experimental results show the advantage of our approach over the current art. In particular, unlike methods tailored for specific applications through training, our approach is applied at inference time, requiring no training. Our webpage is at https://monkeyseedocg.github.io.

Amit H. Bermano, Rinon Gal, Yuval Alaluf et al.

Generative Adversarial Networks (GANs) have established themselves as a prevalent approach to image synthesis. Of these, StyleGAN offers a fascinating case study, owing to its remarkable visual quality and an ability to support a large array of downstream tasks. This state-of-the-art report covers the StyleGAN architecture, and the ways it has been employed since its conception, while also analyzing its severe limitations. It aims to be of use for both newcomers, who wish to get a grasp of the field, and for more experienced readers that might benefit from seeing current research trends and existing tools laid out. Among StyleGAN's most interesting aspects is its learned latent space. Despite being learned with no supervision, it is surprisingly well-behaved and remarkably disentangled. Combined with StyleGAN's visual quality, these properties gave rise to unparalleled editing capabilities. However, the control offered by StyleGAN is inherently limited to the generator's learned distribution, and can only be applied to images generated by StyleGAN itself. Seeking to bring StyleGAN's latent control to real-world scenarios, the study of GAN inversion and latent space embedding has quickly gained in popularity. Meanwhile, this same study has helped shed light on the inner workings and limitations of StyleGAN. We map out StyleGAN's impressive story through these investigations, and discuss the details that have made StyleGAN the go-to generator. We further elaborate on the visual priors StyleGAN constructs, and discuss their use in downstream discriminative tasks. Looking forward, we point out StyleGAN's limitations and speculate on current trends and promising directions for future research, such as task and target specific fine-tuning.

Rotem Tzaban, Ron Mokady, Rinon Gal et al.

The ability of Generative Adversarial Networks to encode rich semantics within their latent space has been widely adopted for facial image editing. However, replicating their success with videos has proven challenging. Sets of high-quality facial videos are lacking, and working with videos introduces a fundamental barrier to overcome - temporal coherency. We propose that this barrier is largely artificial. The source video is already temporally coherent, and deviations from this state arise in part due to careless treatment of individual components in the editing pipeline. We leverage the natural alignment of StyleGAN and the tendency of neural networks to learn low frequency functions, and demonstrate that they provide a strongly consistent prior. We draw on these insights and propose a framework for semantic editing of faces in videos, demonstrating significant improvements over the current state-of-the-art. Our method produces meaningful face manipulations, maintains a higher degree of temporal consistency, and can be applied to challenging, high quality, talking head videos which current methods struggle with.

Dvir Yerushalmi, Dov Danon, Amit H. Bermano

Supervision for image-to-image translation (I2I) tasks is hard to come by, but bears significant effect on the resulting quality. In this paper, we observe that for many Unsupervised I2I (UI2I) scenarios, one domain is more familiar than the other, and offers in-domain prior knowledge, such as semantic segmentation. We argue that for complex scenes, figuring out the semantic structure of the domain is hard, especially with no supervision, but is an important part of a successful I2I operation. We hence introduce two techniques to incorporate this invaluable in-domain prior knowledge for the benefit of translation quality: through a novel Multi-Stream generator architecture, and through a semantic segmentation-based regularization loss term. In essence, we propose splitting the input data according to semantic masks, explicitly guiding the network to different behavior for the different regions of the image. In addition, we propose training a semantic segmentation network along with the translation task, and to leverage this output as a loss term that improves robustness. We validate our approach on urban data, demonstrating superior quality in the challenging UI2I tasks of converting day images to night ones. In addition, we also demonstrate how reinforcing the target dataset with our augmented images improves the training of downstream tasks such as the classical detection one.

Yuval Alaluf, Omer Tov, Ron Mokady et al.

The inversion of real images into StyleGAN's latent space is a well-studied problem. Nevertheless, applying existing approaches to real-world scenarios remains an open challenge, due to an inherent trade-off between reconstruction and editability: latent space regions which can accurately represent real images typically suffer from degraded semantic control. Recent work proposes to mitigate this trade-off by fine-tuning the generator to add the target image to well-behaved, editable regions of the latent space. While promising, this fine-tuning scheme is impractical for prevalent use as it requires a lengthy training phase for each new image. In this work, we introduce this approach into the realm of encoder-based inversion. We propose HyperStyle, a hypernetwork that learns to modulate StyleGAN's weights to faithfully express a given image in editable regions of the latent space. A naive modulation approach would require training a hypernetwork with over three billion parameters. Through careful network design, we reduce this to be in line with existing encoders. HyperStyle yields reconstructions comparable to those of optimization techniques with the near real-time inference capabilities of encoders. Lastly, we demonstrate HyperStyle's effectiveness on several applications beyond the inversion task, including the editing of out-of-domain images which were never seen during training.

Ron Mokady, Rotem Tzaban, Sagie Benaim et al.

The task of unsupervised motion retargeting in videos has seen substantial advancements through the use of deep neural networks. While early works concentrated on specific object priors such as a human face or body, recent work considered the unsupervised case. When the source and target videos, however, are of different shapes, current methods fail. To alleviate this problem, we introduce JOKR - a JOint Keypoint Representation that captures the motion common to both the source and target videos, without requiring any object prior or data collection. By employing a domain confusion term, we enforce the unsupervised keypoint representations of both videos to be indistinguishable. This encourages disentanglement between the parts of the motion that are common to the two domains, and their distinctive appearance and motion, enabling the generation of videos that capture the motion of the one while depicting the style of the other. To enable cases where the objects are of different proportions or orientations, we apply a learned affine transformation between the JOKRs. This augments the representation to be affine invariant, and in practice broadens the variety of possible retargeting pairs. This geometry-driven representation enables further intuitive control, such as temporal coherence and manual editing. Through comprehensive experimentation, we demonstrate the applicability of our method to different challenging cross-domain video pairs. We evaluate our method both qualitatively and quantitatively, and demonstrate that our method handles various cross-domain scenarios, such as different animals, different flowers, and humans. We also demonstrate superior temporal coherency and visual quality compared to state-of-the-art alternatives, through statistical metrics and a user study. Source code and videos can be found at https://rmokady.github.io/JOKR/ .

Daniel Roich, Ron Mokady, Amit H. Bermano et al.

Recently, a surge of advanced facial editing techniques have been proposed that leverage the generative power of a pre-trained StyleGAN. To successfully edit an image this way, one must first project (or invert) the image into the pre-trained generator's domain. As it turns out, however, StyleGAN's latent space induces an inherent tradeoff between distortion and editability, i.e. between maintaining the original appearance and convincingly altering some of its attributes. Practically, this means it is still challenging to apply ID-preserving facial latent-space editing to faces which are out of the generator's domain. In this paper, we present an approach to bridge this gap. Our technique slightly alters the generator, so that an out-of-domain image is faithfully mapped into an in-domain latent code. The key idea is pivotal tuning - a brief training process that preserves the editing quality of an in-domain latent region, while changing its portrayed identity and appearance. In Pivotal Tuning Inversion (PTI), an initial inverted latent code serves as a pivot, around which the generator is fined-tuned. At the same time, a regularization term keeps nearby identities intact, to locally contain the effect. This surgical training process ends up altering appearance features that represent mostly identity, without affecting editing capabilities. We validate our technique through inversion and editing metrics, and show preferable scores to state-of-the-art methods. We further qualitatively demonstrate our technique by applying advanced edits (such as pose, age, or expression) to numerous images of well-known and recognizable identities. Finally, we demonstrate resilience to harder cases, including heavy make-up, elaborate hairstyles and/or headwear, which otherwise could not have been successfully inverted and edited by state-of-the-art methods.

Amir Barda, Yotam Erel, Amit H. Bermano

Mesh-based learning is one of the popular approaches nowadays to learn shapes. The most established backbone in this field is MeshCNN. In this paper, we propose infusing MeshCNN with geometric reasoning to achieve higher quality learning. Through careful analysis of the way geometry is represented through-out the network, we submit that this representation should be rigid motion invariant, and should allow reconstructing the original geometry. Accordingly, we introduce the first and second fundamental forms as an edge-centric, rotation and translation invariant, reconstructable representation. In addition, we update the originally proposed pooling scheme to be more geometrically driven. We validate our analysis through experimentation, and present consistent improvement upon the MeshCNN baseline, as well as other more elaborate state-of-the-art architectures. Furthermore, we demonstrate this fundamental forms-based representation opens the door to accessible generative machine learning over meshes.

Felix Petersen, Amit H. Bermano, Oliver Deussen et al.

The long-coveted task of reconstructing 3D geometry from images is still a standing problem. In this paper, we build on the power of neural networks and introduce Pix2Vex, a network trained to convert camera-captured images into 3D geometry. We present a novel differentiable renderer ($DR$) as a forward validation means during training. Our key insight is that $DR$s produce images of a particular appearance, different from typical input images. Hence, we propose adding an image-to-image translation component, converting between these rendering styles. This translation closes the training loop, while allowing to use minimal supervision only, without needing any 3D model as ground truth. Unlike state-of-the-art methods, our $DR$ is $C^\infty$ smooth and thus does not display any discontinuities at occlusions or dis-occlusions. Through our novel training scheme, our network can train on different types of images, where previous work can typically only train on images of a similar appearance to those rendered by a $DR$.