Eli Shechtman

Kai Zhang, Nick Kolkin, Sai Bi et al. · deepmind

We present a method for transferring the artistic features of an arbitrary style image to a 3D scene. Previous methods that perform 3D stylization on point clouds or meshes are sensitive to geometric reconstruction errors for complex real-world scenes. Instead, we propose to stylize the more robust radiance field representation. We find that the commonly used Gram matrix-based loss tends to produce blurry results without faithful brushstrokes, and introduce a nearest neighbor-based loss that is highly effective at capturing style details while maintaining multi-view consistency. We also propose a novel deferred back-propagation method to enable optimization of memory-intensive radiance fields using style losses defined on full-resolution rendered images. Our extensive evaluation demonstrates that our method outperforms baselines by generating artistic appearance that more closely resembles the style image. Please check our project page for video results and open-source implementations: https://www.cs.cornell.edu/projects/arf/ .

Haitian Zheng, Zhe Lin, Jingwan Lu et al.

Recent image inpainting methods have made great progress but often struggle to generate plausible image structures when dealing with large holes in complex images. This is partially due to the lack of effective network structures that can capture both the long-range dependency and high-level semantics of an image. We propose cascaded modulation GAN (CM-GAN), a new network design consisting of an encoder with Fourier convolution blocks that extract multi-scale feature representations from the input image with holes and a dual-stream decoder with a novel cascaded global-spatial modulation block at each scale level. In each decoder block, global modulation is first applied to perform coarse and semantic-aware structure synthesis, followed by spatial modulation to further adjust the feature map in a spatially adaptive fashion. In addition, we design an object-aware training scheme to prevent the network from hallucinating new objects inside holes, fulfilling the needs of object removal tasks in real-world scenarios. Extensive experiments are conducted to show that our method significantly outperforms existing methods in both quantitative and qualitative evaluation. Please refer to the project page: \url{https://github.com/htzheng/CM-GAN-Inpainting}.

Dave Epstein, Taesung Park, Richard Zhang et al. · berkeley

We propose an unsupervised, mid-level representation for a generative model of scenes. The representation is mid-level in that it is neither per-pixel nor per-image; rather, scenes are modeled as a collection of spatial, depth-ordered "blobs" of features. Blobs are differentiably placed onto a feature grid that is decoded into an image by a generative adversarial network. Due to the spatial uniformity of blobs and the locality inherent to convolution, our network learns to associate different blobs with different entities in a scene and to arrange these blobs to capture scene layout. We demonstrate this emergent behavior by showing that, despite training without any supervision, our method enables applications such as easy manipulation of objects within a scene (e.g., moving, removing, and restyling furniture), creation of feasible scenes given constraints (e.g., plausible rooms with drawers at a particular location), and parsing of real-world images into constituent parts. On a challenging multi-category dataset of indoor scenes, BlobGAN outperforms StyleGAN2 in image quality as measured by FID. See our project page for video results and interactive demo: https://www.dave.ml/blobgan

Nupur Kumari, Bingliang Zhang, Richard Zhang et al.

While generative models produce high-quality images of concepts learned from a large-scale database, a user often wishes to synthesize instantiations of their own concepts (for example, their family, pets, or items). Can we teach a model to quickly acquire a new concept, given a few examples? Furthermore, can we compose multiple new concepts together? We propose Custom Diffusion, an efficient method for augmenting existing text-to-image models. We find that only optimizing a few parameters in the text-to-image conditioning mechanism is sufficiently powerful to represent new concepts while enabling fast tuning (~6 minutes). Additionally, we can jointly train for multiple concepts or combine multiple fine-tuned models into one via closed-form constrained optimization. Our fine-tuned model generates variations of multiple new concepts and seamlessly composes them with existing concepts in novel settings. Our method outperforms or performs on par with several baselines and concurrent works in both qualitative and quantitative evaluations while being memory and computationally efficient.

Minguk Kang, Jun-Yan Zhu, Richard Zhang et al.

The recent success of text-to-image synthesis has taken the world by storm and captured the general public's imagination. From a technical standpoint, it also marked a drastic change in the favored architecture to design generative image models. GANs used to be the de facto choice, with techniques like StyleGAN. With DALL-E 2, auto-regressive and diffusion models became the new standard for large-scale generative models overnight. This rapid shift raises a fundamental question: can we scale up GANs to benefit from large datasets like LAION? We find that naÏvely increasing the capacity of the StyleGAN architecture quickly becomes unstable. We introduce GigaGAN, a new GAN architecture that far exceeds this limit, demonstrating GANs as a viable option for text-to-image synthesis. GigaGAN offers three major advantages. First, it is orders of magnitude faster at inference time, taking only 0.13 seconds to synthesize a 512px image. Second, it can synthesize high-resolution images, for example, 16-megapixel pixels in 3.66 seconds. Finally, GigaGAN supports various latent space editing applications such as latent interpolation, style mixing, and vector arithmetic operations.

Nupur Kumari, Bingliang Zhang, Sheng-Yu Wang et al.

Large-scale text-to-image diffusion models can generate high-fidelity images with powerful compositional ability. However, these models are typically trained on an enormous amount of Internet data, often containing copyrighted material, licensed images, and personal photos. Furthermore, they have been found to replicate the style of various living artists or memorize exact training samples. How can we remove such copyrighted concepts or images without retraining the model from scratch? To achieve this goal, we propose an efficient method of ablating concepts in the pretrained model, i.e., preventing the generation of a target concept. Our algorithm learns to match the image distribution for a target style, instance, or text prompt we wish to ablate to the distribution corresponding to an anchor concept. This prevents the model from generating target concepts given its text condition. Extensive experiments show that our method can successfully prevent the generation of the ablated concept while preserving closely related concepts in the model.

Mang Tik Chiu, Xuaner Zhang, Zijun Wei et al. · gatech

Wires and powerlines are common visual distractions that often undermine the aesthetics of photographs. The manual process of precisely segmenting and removing them is extremely tedious and may take up hours, especially on high-resolution photos where wires may span the entire space. In this paper, we present an automatic wire clean-up system that eases the process of wire segmentation and removal/inpainting to within a few seconds. We observe several unique challenges: wires are thin, lengthy, and sparse. These are rare properties of subjects that common segmentation tasks cannot handle, especially in high-resolution images. We thus propose a two-stage method that leverages both global and local contexts to accurately segment wires in high-resolution images efficiently, and a tile-based inpainting strategy to remove the wires given our predicted segmentation masks. We also introduce the first wire segmentation benchmark dataset, WireSegHR. Finally, we demonstrate quantitatively and qualitatively that our wire clean-up system enables fully automated wire removal with great generalization to various wire appearances.

Lucy Chai, Michael Gharbi, Eli Shechtman et al.

Generative models operate at fixed resolution, even though natural images come in a variety of sizes. As high-resolution details are downsampled away and low-resolution images are discarded altogether, precious supervision is lost. We argue that every pixel matters and create datasets with variable-size images, collected at their native resolutions. To take advantage of varied-size data, we introduce continuous-scale training, a process that samples patches at random scales to train a new generator with variable output resolutions. First, conditioning the generator on a target scale allows us to generate higher resolution images than previously possible, without adding layers to the model. Second, by conditioning on continuous coordinates, we can sample patches that still obey a consistent global layout, which also allows for scalable training at higher resolutions. Controlled FFHQ experiments show that our method can take advantage of multi-resolution training data better than discrete multi-scale approaches, achieving better FID scores and cleaner high-frequency details. We also train on other natural image domains including churches, mountains, and birds, and demonstrate arbitrary scale synthesis with both coherent global layouts and realistic local details, going beyond 2K resolution in our experiments. Our project page is available at: https://chail.github.io/anyres-gan/.

Anna Frühstück, Krishna Kumar Singh, Eli Shechtman et al.

While GANs can produce photo-realistic images in ideal conditions for certain domains, the generation of full-body human images remains difficult due to the diversity of identities, hairstyles, clothing, and the variance in pose. Instead of modeling this complex domain with a single GAN, we propose a novel method to combine multiple pretrained GANs, where one GAN generates a global canvas (e.g., human body) and a set of specialized GANs, or insets, focus on different parts (e.g., faces, shoes) that can be seamlessly inserted onto the global canvas. We model the problem as jointly exploring the respective latent spaces such that the generated images can be combined, by inserting the parts from the specialized generators onto the global canvas, without introducing seams. We demonstrate the setup by combining a full body GAN with a dedicated high-quality face GAN to produce plausible-looking humans. We evaluate our results with quantitative metrics and user studies.

Ke Wang, Michaël Gharbi, He Zhang et al.

Learning-based image harmonization techniques are usually trained to undo synthetic random global transformations applied to a masked foreground in a single ground truth photo. This simulated data does not model many of the important appearance mismatches (illumination, object boundaries, etc.) between foreground and background in real composites, leading to models that do not generalize well and cannot model complex local changes. We propose a new semi-supervised training strategy that addresses this problem and lets us learn complex local appearance harmonization from unpaired real composites, where foreground and background come from different images. Our model is fully parametric. It uses RGB curves to correct the global colors and tone and a shading map to model local variations. Our method outperforms previous work on established benchmarks and real composites, as shown in a user study, and processes high-resolution images interactively.

Yotam Nitzan, Michaël Gharbi, Richard Zhang et al.

Can one inject new concepts into an already trained generative model, while respecting its existing structure and knowledge? We propose a new task - domain expansion - to address this. Given a pretrained generator and novel (but related) domains, we expand the generator to jointly model all domains, old and new, harmoniously. First, we note the generator contains a meaningful, pretrained latent space. Is it possible to minimally perturb this hard-earned representation, while maximally representing the new domains? Interestingly, we find that the latent space offers unused, "dormant" directions, which do not affect the output. This provides an opportunity: By "repurposing" these directions, we can represent new domains without perturbing the original representation. In fact, we find that pretrained generators have the capacity to add several - even hundreds - of new domains! Using our expansion method, one "expanded" model can supersede numerous domain-specific models, without expanding the model size. Additionally, a single expanded generator natively supports smooth transitions between domains, as well as composition of domains. Code and project page available at https://yotamnitzan.github.io/domain-expansion/.

Yichen Sheng, Yifan Liu, Jianming Zhang et al.

Shadows are essential for realistic image compositing. Physics-based shadow rendering methods require 3D geometries, which are not always available. Deep learning-based shadow synthesis methods learn a mapping from the light information to an object's shadow without explicitly modeling the shadow geometry. Still, they lack control and are prone to visual artifacts. We introduce pixel heigh, a novel geometry representation that encodes the correlations between objects, ground, and camera pose. The pixel height can be calculated from 3D geometries, manually annotated on 2D images, and can also be predicted from a single-view RGB image by a supervised approach. It can be used to calculate hard shadows in a 2D image based on the projective geometry, providing precise control of the shadows' direction and shape. Furthermore, we propose a data-driven soft shadow generator to apply softness to a hard shadow based on a softness input parameter. Qualitative and quantitative evaluations demonstrate that the proposed pixel height significantly improves the quality of the shadow generation while allowing for controllability.

Nicholas Kolkin, Michal Kucera, Sylvain Paris et al.

We propose Neural Neighbor Style Transfer (NNST), a pipeline that offers state-of-the-art quality, generalization, and competitive efficiency for artistic style transfer. Our approach is based on explicitly replacing neural features extracted from the content input (to be stylized) with those from a style exemplar, then synthesizing the final output based on these rearranged features. While the spirit of our approach is similar to prior work, we show that our design decisions dramatically improve the final visual quality.

S. Mahdi H. Miangoleh, Zoya Bylinskii, Eric Kee et al.

Common editing operations performed by professional photographers include the cleanup operations: de-emphasizing distracting elements and enhancing subjects. These edits are challenging, requiring a delicate balance between manipulating the viewer's attention while maintaining photo realism. While recent approaches can boast successful examples of attention attenuation or amplification, most of them also suffer from frequent unrealistic edits. We propose a realism loss for saliency-guided image enhancement to maintain high realism across varying image types, while attenuating distractors and amplifying objects of interest. Evaluations with professional photographers confirm that we achieve the dual objective of realism and effectiveness, and outperform the recent approaches on their own datasets, while requiring a smaller memory footprint and runtime. We thus offer a viable solution for automating image enhancement and photo cleanup operations.

Tianwei Yin, Michaël Gharbi, Richard Zhang et al.

Diffusion models generate high-quality images but require dozens of forward passes. We introduce Distribution Matching Distillation (DMD), a procedure to transform a diffusion model into a one-step image generator with minimal impact on image quality. We enforce the one-step image generator match the diffusion model at distribution level, by minimizing an approximate KL divergence whose gradient can be expressed as the difference between 2 score functions, one of the target distribution and the other of the synthetic distribution being produced by our one-step generator. The score functions are parameterized as two diffusion models trained separately on each distribution. Combined with a simple regression loss matching the large-scale structure of the multi-step diffusion outputs, our method outperforms all published few-step diffusion approaches, reaching 2.62 FID on ImageNet 64x64 and 11.49 FID on zero-shot COCO-30k, comparable to Stable Diffusion but orders of magnitude faster. Utilizing FP16 inference, our model generates images at 20 FPS on modern hardware.

Dan Ruta, Gemma Canet Tarrés, Andrew Gilbert et al.

Neural Style Transfer (NST) is the field of study applying neural techniques to modify the artistic appearance of a content image to match the style of a reference style image. Traditionally, NST methods have focused on texture-based image edits, affecting mostly low level information and keeping most image structures the same. However, style-based deformation of the content is desirable for some styles, especially in cases where the style is abstract or the primary concept of the style is in its deformed rendition of some content. With the recent introduction of diffusion models, such as Stable Diffusion, we can access far more powerful image generation techniques, enabling new possibilities. In our work, we propose using this new class of models to perform style transfer while enabling deformable style transfer, an elusive capability in previous models. We show how leveraging the priors of these models can expose new artistic controls at inference time, and we document our findings in exploring this new direction for the field of style transfer.

Yuheng Li, Yijun Li, Jingwan Lu et al.

We introduce a new method for diverse foreground generation with explicit control over various factors. Existing image inpainting based foreground generation methods often struggle to generate diverse results and rarely allow users to explicitly control specific factors of variation (e.g., varying the facial identity or expression for face inpainting results). We leverage contrastive learning with latent codes to generate diverse foreground results for the same masked input. Specifically, we define two sets of latent codes, where one controls a pre-defined factor (``known''), and the other controls the remaining factors (``unknown''). The sampled latent codes from the two sets jointly bi-modulate the convolution kernels to guide the generator to synthesize diverse results. Experiments demonstrate the superiority of our method over state-of-the-arts in result diversity and generation controllability.

Lingzhi Zhang, Connelly Barnes, Kevin Wampler et al.

Recently, deep models have established SOTA performance for low-resolution image inpainting, but they lack fidelity at resolutions associated with modern cameras such as 4K or more, and for large holes. We contribute an inpainting benchmark dataset of photos at 4K and above representative of modern sensors. We demonstrate a novel framework that combines deep learning and traditional methods. We use an existing deep inpainting model LaMa to fill the hole plausibly, establish three guide images consisting of structure, segmentation, depth, and apply a multiply-guided PatchMatch to produce eight candidate upsampled inpainted images. Next, we feed all candidate inpaintings through a novel curation module that chooses a good inpainting by column summation on an 8x8 antisymmetric pairwise preference matrix. Our framework's results are overwhelmingly preferred by users over 8 strong baselines, with improvements of quantitative metrics up to 7.4 over the best baseline LaMa, and our technique when paired with 4 different SOTA inpainting backbones improves each such that ours is overwhelmingly preferred by users over a strong super-res baseline.

Haitian Zheng, Zhe Lin, Jingwan Lu et al.

Structure-guided image completion aims to inpaint a local region of an image according to an input guidance map from users. While such a task enables many practical applications for interactive editing, existing methods often struggle to hallucinate realistic object instances in complex natural scenes. Such a limitation is partially due to the lack of semantic-level constraints inside the hole region as well as the lack of a mechanism to enforce realistic object generation. In this work, we propose a learning paradigm that consists of semantic discriminators and object-level discriminators for improving the generation of complex semantics and objects. Specifically, the semantic discriminators leverage pretrained visual features to improve the realism of the generated visual concepts. Moreover, the object-level discriminators take aligned instances as inputs to enforce the realism of individual objects. Our proposed scheme significantly improves the generation quality and achieves state-of-the-art results on various tasks, including segmentation-guided completion, edge-guided manipulation and panoptically-guided manipulation on Places2 datasets. Furthermore, our trained model is flexible and can support multiple editing use cases, such as object insertion, replacement, removal and standard inpainting. In particular, our trained model combined with a novel automatic image completion pipeline achieves state-of-the-art results on the standard inpainting task.

Dan Ruta, Andrew Gilbert, John Collomosse et al.

Style transfer is the task of reproducing the semantic contents of a source image in the artistic style of a second target image. In this paper, we present NeAT, a new state-of-the art feed-forward style transfer method. We re-formulate feed-forward style transfer as image editing, rather than image generation, resulting in a model which improves over the state-of-the-art in both preserving the source content and matching the target style. An important component of our model's success is identifying and fixing "style halos", a commonly occurring artefact across many style transfer techniques. In addition to training and testing on standard datasets, we introduce the BBST-4M dataset, a new, large scale, high resolution dataset of 4M images. As a component of curating this data, we present a novel model able to classify if an image is stylistic. We use BBST-4M to improve and measure the generalization of NeAT across a huge variety of styles. Not only does NeAT offer state-of-the-art quality and generalization, it is designed and trained for fast inference at high resolution.

Lingzhi Zhang, Zhengjie Xu, Connelly Barnes et al.

Recent advancements in deep generative models have facilitated the creation of photo-realistic images across various tasks. However, these generated images often exhibit perceptual artifacts in specific regions, necessitating manual correction. In this study, we present a comprehensive empirical examination of Perceptual Artifacts Localization (PAL) spanning diverse image synthesis endeavors. We introduce a novel dataset comprising 10,168 generated images, each annotated with per-pixel perceptual artifact labels across ten synthesis tasks. A segmentation model, trained on our proposed dataset, effectively localizes artifacts across a range of tasks. Additionally, we illustrate its proficiency in adapting to previously unseen models using minimal training samples. We further propose an innovative zoom-in inpainting pipeline that seamlessly rectifies perceptual artifacts in the generated images. Through our experimental analyses, we elucidate several practical downstream applications, such as automated artifact rectification, non-referential image quality evaluation, and abnormal region detection in images. The dataset and code are released.

Xiaodan Du, Raymond A. Yeh, Nicholas Kolkin et al.

We propose Fast text2StyleGAN, a natural language interface that adapts pre-trained GANs for text-guided human face synthesis. Leveraging the recent advances in Contrastive Language-Image Pre-training (CLIP), no text data is required during training. Fast text2StyleGAN is formulated as a conditional variational autoencoder (CVAE) that provides extra control and diversity to the generated images at test time. Our model does not require re-training or fine-tuning of the GANs or CLIP when encountering new text prompts. In contrast to prior work, we do not rely on optimization at test time, making our method orders of magnitude faster than prior work. Empirically, on FFHQ dataset, our method offers faster and more accurate generation of images from natural language descriptions with varying levels of detail compared to prior work.

Jaskirat Singh, Boyang Zheng, Zongze Wu et al.

Representation Autoencoders (RAE) replace traditional VAE with pretrained vision encoders. In this paper, we systematically investigate several design choices and find three insights which simplify and improve RAE. First, we study a generalized formulation where the representation is defined as sum of the last k encoder layers rather than solely the final layer. This simple change greatly improves reconstruction without encoder finetuning or specialized data (e.g., text, faces). Second, we study the prevalent assumption that RAE (using pretrained representation as encoder) replaces representation alignment (REPA), which distills the same representation to intermediate layers instead. Through large-scale empirical analysis, we uncover a surprising finding: RAE and REPA exhibit complementary working mechanisms, allowing the same representation to be used as both encoder and target for intermediate diffusion layers. Finally, the original RAE struggles with classifier-free guidance (CFG) and requires training a second, weaker diffusion model for AutoGuidance (AG). We show that REPA itself can be viewed as x-prediction in RAE latent space. By simply re-parameterizing the output of the DiT model, it can provide guidance for "free". Overall, RAEv2 leads to more than 10x faster convergence over the original RAE, achieving a state-of-the-art gFID of 1.06 in just 80 epochs on ImageNet-256. On FDr^k, RAEv2 achieves a state-of-the-art 2.17 at just 80 epochs compared to the previous best 3.26 (800 epochs) without any post-training. This motivates EP_FID@k (epochs to reach unguided gFID <= k) as a measure of training efficiency. RAEv2 attains an EP_FID@2 of 35 epochs, versus 177 for the original RAE. We also validate our approach across diverse settings for text-to-image generation and navigation world models, showing consistent improvements. Code is available at https://raev2.github.io.

Zongze Wu, Nicholas Kolkin, Jonathan Brandt et al.

We address the challenges of precise image inversion and disentangled image editing in the context of few-step diffusion models. We introduce an encoder based iterative inversion technique. The inversion network is conditioned on the input image and the reconstructed image from the previous step, allowing for correction of the next reconstruction towards the input image. We demonstrate that disentangled controls can be easily achieved in the few-step diffusion model by conditioning on an (automatically generated) detailed text prompt. To manipulate the inverted image, we freeze the noise maps and modify one attribute in the text prompt (either manually or via instruction based editing driven by an LLM), resulting in the generation of a new image similar to the input image with only one attribute changed. It can further control the editing strength and accept instructive text prompt. Our approach facilitates realistic text-guided image edits in real-time, requiring only 8 number of functional evaluations (NFEs) in inversion (one-time cost) and 4 NFEs per edit. Our method is not only fast, but also significantly outperforms state-of-the-art multi-step diffusion editing techniques.

ShahRukh Athar, Zexiang Xu, Kalyan Sunkavalli et al.

Volumetric neural rendering methods, such as neural radiance fields (NeRFs), have enabled photo-realistic novel view synthesis. However, in their standard form, NeRFs do not support the editing of objects, such as a human head, within a scene. In this work, we propose RigNeRF, a system that goes beyond just novel view synthesis and enables full control of head pose and facial expressions learned from a single portrait video. We model changes in head pose and facial expressions using a deformation field that is guided by a 3D morphable face model (3DMM). The 3DMM effectively acts as a prior for RigNeRF that learns to predict only residuals to the 3DMM deformations and allows us to render novel (rigid) poses and (non-rigid) expressions that were not present in the input sequence. Using only a smartphone-captured short video of a subject for training, we demonstrate the effectiveness of our method on free view synthesis of a portrait scene with explicit head pose and expression controls. The project page can be found here: http://shahrukhathar.github.io/2022/06/06/RigNeRF.html

Yicong Hong, Yiqun Mei, Chongjian Ge et al.

A truly interactive world model requires three key ingredients: real-time long-horizon streaming, consistent spatial memory, and precise user control. However, most existing approaches address only one of these aspects in isolation, as achieving all three simultaneously is highly challenging-for example, long-term memory mechanisms often degrade real-time performance. In this work, we present RELIC, a unified framework that tackles these three challenges altogether. Given a single image and a text description, RELIC enables memory-aware, long-duration exploration of arbitrary scenes in real time. Built upon recent autoregressive video-diffusion distillation techniques, our model represents long-horizon memory using highly compressed historical latent tokens encoded with both relative actions and absolute camera poses within the KV cache. This compact, camera-aware memory structure supports implicit 3D-consistent content retrieval and enforces long-term coherence with minimal computational overhead. In parallel, we fine-tune a bidirectional teacher video model to generate sequences beyond its original 5-second training horizon, and transform it into a causal student generator using a new memory-efficient self-forcing paradigm that enables full-context distillation over long-duration teacher as well as long student self-rollouts. Implemented as a 14B-parameter model and trained on a curated Unreal Engine-rendered dataset, RELIC achieves real-time generation at 16 FPS while demonstrating more accurate action following, more stable long-horizon streaming, and more robust spatial-memory retrieval compared with prior work. These capabilities establish RELIC as a strong foundation for the next generation of interactive world modeling.

Jaskirat Singh, Xingjian Leng, Zongze Wu et al.

Representation alignment (REPA) guides generative training by distilling representations from a strong, pretrained vision encoder to intermediate diffusion features. We investigate a fundamental question: what aspect of the target representation matters for generation, its \textit{global} \revision{semantic} information (e.g., measured by ImageNet-1K accuracy) or its spatial structure (i.e. pairwise cosine similarity between patch tokens)? Prevalent wisdom holds that stronger global semantic performance leads to better generation as a target representation. To study this, we first perform a large-scale empirical analysis across 27 different vision encoders and different model scales. The results are surprising; spatial structure, rather than global performance, drives the generation performance of a target representation. To further study this, we introduce two straightforward modifications, which specifically accentuate the transfer of \emph{spatial} information. We replace the standard MLP projection layer in REPA with a simple convolution layer and introduce a spatial normalization layer for the external representation. Surprisingly, our simple method (implemented in $<$4 lines of code), termed iREPA, consistently improves convergence speed of REPA, across a diverse set of vision encoders, model sizes, and training variants (such as REPA, REPA-E, Meanflow, JiT etc). %, etc. Our work motivates revisiting the fundamental working mechanism of representational alignment and how it can be leveraged for improved training of generative models. The code and project page are available at https://end2end-diffusion.github.io/irepa

Xin Yu, Xiaojuan Qi, Zhengqi Li et al.

We introduce the Self-Evaluating Model (Self-E), a novel, from-scratch training approach for text-to-image generation that supports any-step inference. Self-E learns from data similarly to a Flow Matching model, while simultaneously employing a novel self-evaluation mechanism: it evaluates its own generated samples using its current score estimates, effectively serving as a dynamic self-teacher. Unlike traditional diffusion or flow models, it does not rely solely on local supervision, which typically necessitates many inference steps. Unlike distillation-based approaches, it does not require a pretrained teacher. This combination of instantaneous local learning and self-driven global matching bridges the gap between the two paradigms, enabling the training of a high-quality text-to-image model from scratch that excels even at very low step counts. Extensive experiments on large-scale text-to-image benchmarks show that Self-E not only excels in few-step generation, but is also competitive with state-of-the-art Flow Matching models at 50 steps. We further find that its performance improves monotonically as inference steps increase, enabling both ultra-fast few-step generation and high-quality long-trajectory sampling within a single unified model. To our knowledge, Self-E is the first from-scratch, any-step text-to-image model, offering a unified framework for efficient and scalable generation.

Zhengyang Geng, Yiyang Lu, Zongze Wu et al.

MeanFlow (MF) has recently been established as a framework for one-step generative modeling. However, its ``fastforward'' nature introduces key challenges in both the training objective and the guidance mechanism. First, the original MF's training target depends not only on the underlying ground-truth fields but also on the network itself. To address this issue, we recast the objective as a loss on the instantaneous velocity $v$, re-parameterized by a network that predicts the average velocity $u$. Our reformulation yields a more standard regression problem and improves the training stability. Second, the original MF fixes the classifier-free guidance scale during training, which sacrifices flexibility. We tackle this issue by formulating guidance as explicit conditioning variables, thereby retaining flexibility at test time. The diverse conditions are processed through in-context conditioning, which reduces model size and benefits performance. Overall, our $\textbf{improved MeanFlow}$ ($\textbf{iMF}$) method, trained entirely from scratch, achieves $\textbf{1.72}$ FID with a single function evaluation (1-NFE) on ImageNet 256$\times$256. iMF substantially outperforms prior methods of this kind and closes the gap with multi-step methods while using no distillation. We hope our work will further advance fastforward generative modeling as a stand-alone paradigm.

Haotian Xue, Qi Chen, Zhonghao Wang et al.

Video diffusion models achieve strong frame-level fidelity but still struggle with motion coherence, dynamics and realism, often producing jitter, ghosting, or implausible dynamics. A key limitation is that the standard denoising MSE objective provides no direct supervision on temporal consistency, allowing models to achieve low loss while still generating poor motion. We propose MoGAN, a motion-centric post-training framework that improves motion realism without reward models or human preference data. Built atop a 3-step distilled video diffusion model, we train a DiT-based optical-flow discriminator to differentiate real from generated motion, combined with a distribution-matching regularizer to preserve visual fidelity. With experiments on Wan2.1-T2V-1.3B, MoGAN substantially improves motion quality across benchmarks. On VBench, MoGAN boosts motion score by +7.3% over the 50-step teacher and +13.3% over the 3-step DMD model. On VideoJAM-Bench, MoGAN improves motion score by +7.4% over the teacher and +8.8% over DMD, while maintaining comparable or even better aesthetic and image-quality scores. A human study further confirms that MoGAN is preferred for motion quality (52% vs. 38% for the teacher; 56% vs. 29% for DMD). Overall, MoGAN delivers significantly more realistic motion without sacrificing visual fidelity or efficiency, offering a practical path toward fast, high-quality video generation. Project webpage is: https://xavihart.github.io/mogan.

Sicheng Mo, Thao Nguyen, Richard Zhang et al.

In this work, we explore an untapped signal in diffusion model inference. While all previous methods generate images independently at inference, we instead ask if samples can be generated collaboratively. We propose Group Diffusion, unlocking the attention mechanism to be shared across images, rather than limited to just the patches within an image. This enables images to be jointly denoised at inference time, learning both intra and inter-image correspondence. We observe a clear scaling effect - larger group sizes yield stronger cross-sample attention and better generation quality. Furthermore, we introduce a qualitative measure to capture this behavior and show that its strength closely correlates with FID. Built on standard diffusion transformers, our GroupDiff achieves up to 32.2% FID improvement on ImageNet-256x256. Our work reveals cross-sample inference as an effective, previously unexplored mechanism for generative modeling.

Rohit Gandikota, Zongze Wu, Richard Zhang et al.

We present SliderSpace, a framework for automatically decomposing the visual capabilities of diffusion models into controllable and human-understandable directions. Unlike existing control methods that require a user to specify attributes for each edit direction individually, SliderSpace discovers multiple interpretable and diverse directions simultaneously from a single text prompt. Each direction is trained as a low-rank adaptor, enabling compositional control and the discovery of surprising possibilities in the model's latent space. Through extensive experiments on state-of-the-art diffusion models, we demonstrate SliderSpace's effectiveness across three applications: concept decomposition, artistic style exploration, and diversity enhancement. Our quantitative evaluation shows that SliderSpace-discovered directions decompose the visual structure of model's knowledge effectively, offering insights into the latent capabilities encoded within diffusion models. User studies further validate that our method produces more diverse and useful variations compared to baselines. Our code, data and trained weights are available at https://sliderspace.baulab.info

Xingjian Bai, Guande He, Zhengqi Li et al.

Causality -- referring to temporal, uni-directional cause-effect relationships between components -- underlies many complex generative processes, including videos, language, and robot trajectories. Current causal diffusion models entangle temporal reasoning with iterative denoising, applying causal attention across all layers, at every denoising step, and over the entire context. In this paper, we show that the causal reasoning in these models is separable from the multi-step denoising process. Through systematic probing of autoregressive video diffusers, we uncover two key regularities: (1) early layers produce highly similar features across denoising steps, indicating redundant computation along the diffusion trajectory; and (2) deeper layers exhibit sparse cross-frame attention and primarily perform intra-frame rendering. Motivated by these findings, we introduce Separable Causal Diffusion (SCD), a new architecture that explicitly decouples once-per-frame temporal reasoning, via a causal transformer encoder, from multi-step frame-wise rendering, via a lightweight diffusion decoder. Extensive experiments on both pretraining and post-training tasks across synthetic and real benchmarks show that SCD significantly improves throughput and per-frame latency while matching or surpassing the generation quality of strong causal diffusion baselines.

Tianwei Yin, Michaël Gharbi, Taesung Park et al.

Recent approaches have shown promises distilling diffusion models into efficient one-step generators. Among them, Distribution Matching Distillation (DMD) produces one-step generators that match their teacher in distribution, without enforcing a one-to-one correspondence with the sampling trajectories of their teachers. However, to ensure stable training, DMD requires an additional regression loss computed using a large set of noise-image pairs generated by the teacher with many steps of a deterministic sampler. This is costly for large-scale text-to-image synthesis and limits the student's quality, tying it too closely to the teacher's original sampling paths. We introduce DMD2, a set of techniques that lift this limitation and improve DMD training. First, we eliminate the regression loss and the need for expensive dataset construction. We show that the resulting instability is due to the fake critic not estimating the distribution of generated samples accurately and propose a two time-scale update rule as a remedy. Second, we integrate a GAN loss into the distillation procedure, discriminating between generated samples and real images. This lets us train the student model on real data, mitigating the imperfect real score estimation from the teacher model, and enhancing quality. Lastly, we modify the training procedure to enable multi-step sampling. We identify and address the training-inference input mismatch problem in this setting, by simulating inference-time generator samples during training time. Taken together, our improvements set new benchmarks in one-step image generation, with FID scores of 1.28 on ImageNet-64x64 and 8.35 on zero-shot COCO 2014, surpassing the original teacher despite a 500X reduction in inference cost. Further, we show our approach can generate megapixel images by distilling SDXL, demonstrating exceptional visual quality among few-step methods.

Shivam Duggal, Xingjian Bai, Zongze Wu et al.

Latent diffusion models (LDMs) enable high-fidelity synthesis by operating in learned latent spaces. However, training state-of-the-art LDMs requires complex staging: a tokenizer must be trained first, before the diffusion model can be trained in the frozen latent space. We propose UNITE - an autoencoder architecture for unified tokenization and latent diffusion. UNITE consists of a Generative Encoder that serves as both image tokenizer and latent generator via weight sharing. Our key insight is that tokenization and generation can be viewed as the same latent inference problem under different conditioning regimes: tokenization infers latents from fully observed images, whereas generation infers them from noise together with text or class conditioning. Motivated by this, we introduce a single-stage training procedure that jointly optimizes both tasks via two forward passes through the same Generative Encoder. The shared parameters enable gradients to jointly shape the latent space, encouraging a "common latent language". Across image and molecule modalities, UNITE achieves near state of the art performance without adversarial losses or pretrained encoders (e.g., DINO), reaching FID 2.12 and 1.73 for Base and Large models on ImageNet 256 x 256. We further analyze the Generative Encoder through the lenses of representation alignment and compression. These results show that single stage joint training of tokenization & generation from scratch is feasible.

Yao-Chih Lee, Zhoutong Zhang, Jiahui Huang et al.

Camera and object motions are central to a video's narrative. However, precisely editing these captured motions remains a significant challenge, especially under complex object movements. Current motion-controlled image-to-video (I2V) approaches often lack full-scene context for consistent video editing, while video-to-video (V2V) methods provide viewpoint changes or basic object translation, but offer limited control over fine-grained object motion. We present a track-conditioned V2V framework that enables joint editing of camera and object motion. We achieve this by conditioning a video generation model on a source video and paired 3D point tracks representing source and target motions. These 3D tracks establish sparse correspondences that transfer rich context from the source video to new motions while preserving spatiotemporal coherence. Crucially, compared to 2D tracks, 3D tracks provide explicit depth cues, allowing the model to resolve depth order and handle occlusions for precise motion editing. Trained in two stages on synthetic and real data, our model supports diverse motion edits, including joint camera/object manipulation, motion transfer, and non-rigid deformation, unlocking new creative potential in video editing.

Thao Nguyen, Sicheng Mo, Krishna Kumar Singh et al.

Humans do not just see attribute similarity -- we also see relational similarity. An apple is like a peach because both are reddish fruit, but the Earth is also like a peach: its crust, mantle, and core correspond to the peach's skin, flesh, and pit. This ability to perceive and recognize relational similarity, is arguable by cognitive scientist to be what distinguishes humans from other species. Yet, all widely used visual similarity metrics today (e.g., LPIPS, CLIP, DINO) focus solely on perceptual attribute similarity and fail to capture the rich, often surprising relational similarities that humans perceive. How can we go beyond the visible content of an image to capture its relational properties? How can we bring images with the same relational logic closer together in representation space? To answer these questions, we first formulate relational image similarity as a measurable problem: two images are relationally similar when their internal relations or functions among visual elements correspond, even if their visual attributes differ. We then curate 114k image-caption dataset in which the captions are anonymized -- describing the underlying relational logic of the scene rather than its surface content. Using this dataset, we finetune a Vision-Language model to measure the relational similarity between images. This model serves as the first step toward connecting images by their underlying relational structure rather than their visible appearance. Our study shows that while relational similarity has a lot of real-world applications, existing image similarity models fail to capture it -- revealing a critical gap in visual computing.

Tianwei Yin, Qiang Zhang, Richard Zhang et al.

Current video diffusion models achieve impressive generation quality but struggle in interactive applications due to bidirectional attention dependencies. The generation of a single frame requires the model to process the entire sequence, including the future. We address this limitation by adapting a pretrained bidirectional diffusion transformer to an autoregressive transformer that generates frames on-the-fly. To further reduce latency, we extend distribution matching distillation (DMD) to videos, distilling 50-step diffusion model into a 4-step generator. To enable stable and high-quality distillation, we introduce a student initialization scheme based on teacher's ODE trajectories, as well as an asymmetric distillation strategy that supervises a causal student model with a bidirectional teacher. This approach effectively mitigates error accumulation in autoregressive generation, allowing long-duration video synthesis despite training on short clips. Our model achieves a total score of 84.27 on the VBench-Long benchmark, surpassing all previous video generation models. It enables fast streaming generation of high-quality videos at 9.4 FPS on a single GPU thanks to KV caching. Our approach also enables streaming video-to-video translation, image-to-video, and dynamic prompting in a zero-shot manner.

Xun Huang, Zhengqi Li, Guande He et al.

We introduce Self Forcing, a novel training paradigm for autoregressive video diffusion models. It addresses the longstanding issue of exposure bias, where models trained on ground-truth context must generate sequences conditioned on their own imperfect outputs during inference. Unlike prior methods that denoise future frames based on ground-truth context frames, Self Forcing conditions each frame's generation on previously self-generated outputs by performing autoregressive rollout with key-value (KV) caching during training. This strategy enables supervision through a holistic loss at the video level that directly evaluates the quality of the entire generated sequence, rather than relying solely on traditional frame-wise objectives. To ensure training efficiency, we employ a few-step diffusion model along with a stochastic gradient truncation strategy, effectively balancing computational cost and performance. We further introduce a rolling KV cache mechanism that enables efficient autoregressive video extrapolation. Extensive experiments demonstrate that our approach achieves real-time streaming video generation with sub-second latency on a single GPU, while matching or even surpassing the generation quality of significantly slower and non-causal diffusion models. Project website: http://self-forcing.github.io/

Yiran Xu, Taesung Park, Richard Zhang et al.

Video super-resolution (VSR) approaches have shown impressive temporal consistency in upsampled videos. However, these approaches tend to generate blurrier results than their image counterparts as they are limited in their generative capability. This raises a fundamental question: can we extend the success of a generative image upsampler to the VSR task while preserving the temporal consistency? We introduce VideoGigaGAN, a new generative VSR model that can produce videos with high-frequency details and temporal consistency. VideoGigaGAN builds upon a large-scale image upsampler -- GigaGAN. Simply inflating GigaGAN to a video model by adding temporal modules produces severe temporal flickering. We identify several key issues and propose techniques that significantly improve the temporal consistency of upsampled videos. Our experiments show that, unlike previous VSR methods, VideoGigaGAN generates temporally consistent videos with more fine-grained appearance details. We validate the effectiveness of VideoGigaGAN by comparing it with state-of-the-art VSR models on public datasets and showcasing video results with $8\times$ super-resolution.

Yotam Nitzan, Zongze Wu, Richard Zhang et al.

We introduce a novel diffusion transformer, LazyDiffusion, that generates partial image updates efficiently. Our approach targets interactive image editing applications in which, starting from a blank canvas or an image, a user specifies a sequence of localized image modifications using binary masks and text prompts. Our generator operates in two phases. First, a context encoder processes the current canvas and user mask to produce a compact global context tailored to the region to generate. Second, conditioned on this context, a diffusion-based transformer decoder synthesizes the masked pixels in a "lazy" fashion, i.e., it only generates the masked region. This contrasts with previous works that either regenerate the full canvas, wasting time and computation, or confine processing to a tight rectangular crop around the mask, ignoring the global image context altogether. Our decoder's runtime scales with the mask size, which is typically small, while our encoder introduces negligible overhead. We demonstrate that our approach is competitive with state-of-the-art inpainting methods in terms of quality and fidelity while providing a 10x speedup for typical user interactions, where the editing mask represents 10% of the image.

Joanna Materzynska, Josef Sivic, Eli Shechtman et al.

We introduce an approach for augmenting text-to-video generation models with customized motions, extending their capabilities beyond the motions depicted in the original training data. By leveraging a few video samples demonstrating specific movements as input, our method learns and generalizes the input motion patterns for diverse, text-specified scenarios. Our contributions are threefold. First, to achieve our results, we finetune an existing text-to-video model to learn a novel mapping between the depicted motion in the input examples to a new unique token. To avoid overfitting to the new custom motion, we introduce an approach for regularization over videos. Second, by leveraging the motion priors in a pretrained model, our method can produce novel videos featuring multiple people doing the custom motion, and can invoke the motion in combination with other motions. Furthermore, our approach extends to the multimodal customization of motion and appearance of individualized subjects, enabling the generation of videos featuring unique characters and distinct motions. Third, to validate our method, we introduce an approach for quantitatively evaluating the learned custom motion and perform a systematic ablation study. We show that our method significantly outperforms prior appearance-based customization approaches when extended to the motion customization task.

Jiteng Mu, Michaël Gharbi, Richard Zhang et al.

Diffusion models have made significant advances in text-guided synthesis tasks. However, editing user-provided images remains challenging, as the high dimensional noise input space of diffusion models is not naturally suited for image inversion or spatial editing. In this work, we propose an image representation that promotes spatial editing of input images using a diffusion model. Concretely, we learn to encode an input into "image elements" that can faithfully reconstruct an input image. These elements can be intuitively edited by a user, and are decoded by a diffusion model into realistic images. We show the effectiveness of our representation on various image editing tasks, such as object resizing, rearrangement, dragging, de-occlusion, removal, variation, and image composition. Project page: https://jitengmu.github.io/Editable_Image_Elements/

Nupur Kumari, Grace Su, Richard Zhang et al.

Model customization introduces new concepts to existing text-to-image models, enabling the generation of these new concepts/objects in novel contexts. However, such methods lack accurate camera view control with respect to the new object, and users must resort to prompt engineering (e.g., adding ``top-view'') to achieve coarse view control. In this work, we introduce a new task -- enabling explicit control of the object viewpoint in the customization of text-to-image diffusion models. This allows us to modify the custom object's properties and generate it in various background scenes via text prompts, all while incorporating the object viewpoint as an additional control. This new task presents significant challenges, as one must harmoniously merge a 3D representation from the multi-view images with the 2D pre-trained model. To bridge this gap, we propose to condition the diffusion process on the 3D object features rendered from the target viewpoint. During training, we fine-tune the 3D feature prediction modules to reconstruct the object's appearance and geometry, while reducing overfitting to the input multi-view images. Our method outperforms existing image editing and model customization baselines in preserving the custom object's identity while following the target object viewpoint and the text prompt.

Haoran You, Connelly Barnes, Yuqian Zhou et al.

Diffusion Transformers (DiTs) have achieved state-of-the-art (SOTA) image generation quality but suffer from high latency and memory inefficiency, making them difficult to deploy on resource-constrained devices. One major efficiency bottleneck is that existing DiTs apply equal computation across all regions of an image. However, not all image tokens are equally important, and certain localized areas require more computation, such as objects. To address this, we propose DiffCR, a dynamic DiT inference framework with differentiable compression ratios, which automatically learns to dynamically route computation across layers and timesteps for each image token, resulting in efficient DiTs. Specifically, DiffCR integrates three features: (1) A token-level routing scheme where each DiT layer includes a router that is fine-tuned jointly with model weights to predict token importance scores. In this way, unimportant tokens bypass the entire layer's computation; (2) A layer-wise differentiable ratio mechanism where different DiT layers automatically learn varying compression ratios from a zero initialization, resulting in large compression ratios in redundant layers while others remain less compressed or even uncompressed; (3) A timestep-wise differentiable ratio mechanism where each denoising timestep learns its own compression ratio. The resulting pattern shows higher ratios for noisier timesteps and lower ratios as the image becomes clearer. Extensive experiments on text-to-image and inpainting tasks show that DiffCR effectively captures dynamism across token, layer, and timestep axes, achieving superior trade-offs between generation quality and efficiency compared to prior works. The project website is available at https://www.haoranyou.com/diffcr.

Sicheng Mo, Thao Nguyen, Xun Huang et al.

We propose X-Fusion, a framework that extends pretrained Large Language Models (LLMs) for multimodal tasks while preserving their language capabilities. X-Fusion employs a dual-tower design with modality-specific weights, keeping the LLM's parameters frozen while integrating vision-specific information for both understanding and generation. Our experiments demonstrate that X-Fusion consistently outperforms alternative architectures on both image-to-text and text-to-image tasks. We find that incorporating understanding-focused data improves generation quality, reducing image data noise enhances overall performance, and feature alignment accelerates convergence for smaller models but has minimal impact on larger ones. Our findings provide valuable insights into building efficient unified multimodal models.

Xiaojuan Wang, Taesung Park, Yang Zhou et al.

A jump cut offers an abrupt, sometimes unwanted change in the viewing experience. We present a novel framework for smoothing these jump cuts, in the context of talking head videos. We leverage the appearance of the subject from the other source frames in the video, fusing it with a mid-level representation driven by DensePose keypoints and face landmarks. To achieve motion, we interpolate the keypoints and landmarks between the end frames around the cut. We then use an image translation network from the keypoints and source frames, to synthesize pixels. Because keypoints can contain errors, we propose a cross-modal attention scheme to select and pick the most appropriate source amongst multiple options for each key point. By leveraging this mid-level representation, our method can achieve stronger results than a strong video interpolation baseline. We demonstrate our method on various jump cuts in the talking head videos, such as cutting filler words, pauses, and even random cuts. Our experiments show that we can achieve seamless transitions, even in the challenging cases where the talking head rotates or moves drastically in the jump cut.

Jingdong Zhang, Lingzhi Zhang, Qing Liu et al.

Digital images are often degraded by soft effects such as lens flare, haze, shadows, and reflections, which reduce aesthetics even though the underlying pixels remain partially visible. The prevailing works address these degradations in isolation, developing highly specialized, specialist models that lack scalability and fail to exploit the shared underlying essences of these restoration problems. While specialist models are limited, recent large-scale pretrained generalist models offer powerful, text-driven image editing capabilities. while recent general-purpose systems (e.g., GPT-4o, Flux Kontext, Nano Banana) require detailed prompts and often fail to achieve robust removal on these fine-grained tasks or preserve identity of the scene. Leveraging the common essence of soft effects, i.e., semi-transparent occlusions, we introduce a foundational versatile model UniSER, capable of addressing diverse degradations caused by soft effects within a single framework. Our methodology centers on curating a massive 3.8M-pair dataset to ensure robustness and generalization, which includes novel, physically-plausible data to fill critical gaps in public benchmarks, and a tailored training pipeline that fine-tunes a Diffusion Transformer to learn robust restoration priors from this diverse data, integrating fine-grained mask and strength controls. This synergistic approach allows UniSER to significantly outperform both specialist and generalist models, achieving robust, high-fidelity restoration in the wild.

Nupur Kumari, Sheng-Yu Wang, Nanxuan Zhao et al.

Recent image editing models have achieved impressive results while following natural language editing instructions, but they rely on supervised fine-tuning with large datasets of input-target pairs. This is a critical bottleneck, as such naturally occurring pairs are hard to curate at scale. Current workarounds use synthetic training pairs that leverage the zero-shot capabilities of existing models. However, this can propagate and magnify the artifacts of the pretrained model into the final trained model. In this work, we present a new training paradigm that eliminates the need for paired data entirely. Our approach directly optimizes a few-step diffusion model by unrolling it during training and leveraging feedback from vision-language models (VLMs). For each input and editing instruction, the VLM evaluates if an edit follows the instruction and preserves unchanged content, providing direct gradients for end-to-end optimization. To ensure visual fidelity, we incorporate distribution matching loss (DMD), which constrains generated images to remain within the image manifold learned by pretrained models. We evaluate our method on standard benchmarks and include an extensive ablation study. Without any paired data, our method performs on par with various image editing diffusion models trained on extensive supervised paired data, under the few-step setting. Given the same VLM as the reward model, we also outperform RL-based techniques like Flow-GRPO.

Ayush Shrivastava, Connelly Barnes, Xuaner Zhang et al.

Current text-to-image diffusion models excel at generating diverse, high-quality images, yet they struggle to incorporate fine-grained camera metadata such as precise aperture settings. In this work, we introduce a novel text-to-image diffusion framework that leverages camera metadata, or EXIF data, which is often embedded in image files, with an emphasis on generating controllable lens blur. Our method mimics the physical image formation process by first generating an all-in-focus image, estimating its monocular depth, predicting a plausible focus distance with a novel focus distance transformer, and then forming a defocused image with an existing differentiable lens blur model. Gradients flow backwards through this whole process, allowing us to learn without explicit supervision to generate defocus effects based on content elements and the provided EXIF data. At inference time, this enables precise interactive user control over defocus effects while preserving scene contents, which is not achievable with existing diffusion models. Experimental results demonstrate that our model enables superior fine-grained control without altering the depicted scene.