Humans experience the world through multiple modalities, such as, vision, language, and speech, making it natural to explore the commonality and distinctions among them. In this work, we take a data-driven approach to address this question by analyzing interpretable, monosemantic features extracted from deep multimodal models. Specifically, we investigate CLIP, a prominent visual-language representation model trained on massive image-text pairs. Building on prior research in single-modal interpretability, we develop a set of multi-modal interpretability tools and measures designed to disentangle and analyze features learned from CLIP. Specifically, we introduce the Modality Dominance Score (MDS) to attribute each CLIP feature to a specific modality. We then map CLIP features into a more interpretable space, enabling us to categorize them into three distinct classes: vision features (single-modal), language features (single-modal), and visual-language features (cross-modal). Interestingly, this data-driven categorization closely aligns with human intuitive understandings of different modalities. We further show that this modality decomposition can benefit multiple downstream tasks, including reducing bias in gender detection, generating cross-modal adversarial examples, and enabling modal-specific feature control in text-to-image generation. These results indicate that large-scale multimodal models, when equipped with task-agnostic interpretability tools, can offer valuable insights into the relationships between different data modalities.
Autoregressive (AR) models have recently shown strong performance in image generation, where a critical component is the visual tokenizer (VT) that maps continuous pixel inputs to discrete token sequences. The quality of the VT largely defines the upper bound of AR model performance. However, current discrete VTs fall significantly behind continuous variational autoencoders (VAEs), leading to degraded image reconstructions and poor preservation of details and text. Existing benchmarks focus on end-to-end generation quality, without isolating VT performance. To address this gap, we introduce VTBench, a comprehensive benchmark that systematically evaluates VTs across three core tasks: Image Reconstruction, Detail Preservation, and Text Preservation, and covers a diverse range of evaluation scenarios. We systematically assess state-of-the-art VTs using a set of metrics to evaluate the quality of reconstructed images. Our findings reveal that continuous VAEs produce superior visual representations compared to discrete VTs, particularly in retaining spatial structure and semantic detail. In contrast, the degraded representations produced by discrete VTs often lead to distorted reconstructions, loss of fine-grained textures, and failures in preserving text and object integrity. Furthermore, we conduct experiments on GPT-4o image generation and discuss its potential AR nature, offering new insights into the role of visual tokenization. We release our benchmark and codebase publicly to support further research and call on the community to develop strong, general-purpose open-source VTs.
Data steganography aims to conceal information within visual content, yet existing spatial- and frequency-domain approaches suffer from trade-offs between security, capacity, and perceptual quality. Recent advances in generative models, particularly diffusion models, offer new avenues for adaptive image synthesis, but integrating precise information embedding into the generative process remains challenging. We introduce Shackled Dancing Diffusion, or SD$^2$, a plug-and-play generative steganography method that combines bit-position locking with diffusion sampling injection to enable controllable information embedding within the generative trajectory. SD$^2$ leverages the expressive power of diffusion models to synthesize diverse carrier images while maintaining full message recovery with $100\%$ accuracy. Our method achieves a favorable balance between randomness and constraint, enhancing robustness against steganalysis without compromising image fidelity. Extensive experiments show that SD$^2$ substantially outperforms prior methods in security, embedding capacity, and stability. This algorithm offers new insights into controllable generation and opens promising directions for secure visual communication.
Diffusion models have made remarkable advancements in generating high-quality images from textual descriptions. Recent works like LayerDiffuse have extended the previous single-layer, unified image generation paradigm to transparent image layer generation. However, existing multi-layer generation methods fail to handle the interactions among multiple layers such as rational global layout, physics-plausible contacts and visual effects like shadows and reflections while maintaining high alpha quality. To solve this problem, we propose PSDiffusion, a unified diffusion framework for simultaneous multi-layer text-to-image generation. Our model can automatically generate multi-layer images with one RGB background and multiple RGBA foregrounds through a single feed-forward process. Unlike existing methods that combine multiple tools for post-decomposition or generate layers sequentially and separately, our method introduces a global-layer interactive mechanism that generates layered-images concurrently and collaboratively, ensuring not only high quality and completeness for each layer, but also spatial and visual interactions among layers for global coherence.
The remarkable ease of use of diffusion models for image generation has led to a proliferation of synthetic content online. While these models are often employed for legitimate purposes, they are also used to generate fake images that support misinformation and hate speech. Consequently, it is crucial to develop robust tools capable of detecting whether an image has been generated by such models. Many current detection methods, however, require large volumes of sample images for training. Unfortunately, due to the rapid evolution of the field, existing datasets often cover only a limited range of models and quickly become outdated. In this work, we introduce DRAGON, a comprehensive dataset comprising images from 25 diffusion models, spanning both recent advancements and older, well-established architectures. The dataset contains a broad variety of images representing diverse subjects. To enhance image realism, we propose a simple yet effective pipeline that leverages a large language model to expand input prompts, thereby generating more diverse and higher-quality outputs, as evidenced by improvements in standard quality metrics. The dataset is provided in multiple sizes (ranging from extra-small to extra-large) to accomodate different research scenarios. DRAGON is designed to support the forensic community in developing and evaluating detection and attribution techniques for synthetic content. Additionally, the dataset is accompanied by a dedicated test set, intended to serve as a benchmark for assessing the performance of newly developed methods.
Diffusion models have made substantial advances in image generation, yet models trained on large, unfiltered datasets often yield outputs misaligned with human preferences. Numerous methods have been proposed to fine-tune pre-trained diffusion models, achieving notable improvements in aligning generated outputs with human preferences. However, we argue that existing preference alignment methods neglect the critical role of handling unconditional/negative-conditional outputs, leading to a diminished capacity to avoid generating undesirable outcomes. This oversight limits the efficacy of classifier-free guidance~(CFG), which relies on the contrast between conditional generation and unconditional/negative-conditional generation to optimize output quality. In response, we propose a straightforward but versatile effective approach that involves training a model specifically attuned to negative preferences. This method does not require new training strategies or datasets but rather involves minor modifications to existing techniques. Our approach integrates seamlessly with models such as SD1.5, SDXL, video diffusion models and models that have undergone preference optimization, consistently enhancing their alignment with human preferences.
State-of-the-art T2I models are capable of generating high-resolution images given textual prompts. However, they still struggle with accurately depicting compositional scenes that specify multiple objects, attributes, and spatial relations. We present CompAlign, a challenging benchmark with an emphasis on assessing the depiction of 3D-spatial relationships, for evaluating and improving models on compositional image generation. CompAlign consists of 900 complex multi-subject image generation prompts that combine numerical and 3D-spatial relationships with varied attribute bindings. Our benchmark is remarkably challenging, incorporating generation tasks with 3+ generation subjects with complex 3D-spatial relationships. Additionally, we propose CompQuest, an interpretable and accurate evaluation framework that decomposes complex prompts into atomic sub-questions, then utilizes a MLLM to provide fine-grained binary feedback on the correctness of each aspect of generation elements in model-generated images. This enables precise quantification of alignment between generated images and compositional prompts. Furthermore, we propose an alignment framework that uses CompQuest's feedback as preference signals to improve diffusion models' compositional image generation abilities. Using adjustable per-image preferences, our method is easily scalable and flexible for different tasks. Evaluation of 9 T2I models reveals that: (1) models remarkable struggle more with compositional tasks with more complex 3D-spatial configurations, and (2) a noticeable performance gap exists between open-source accessible models and closed-source commercial models. Further empirical study on using CompAlign for model alignment yield promising results: post-alignment diffusion models achieve remarkable improvements in compositional accuracy, especially on complex generation tasks, outperforming previous approaches.
Concept erasing has recently emerged as an effective paradigm to prevent text-to-image diffusion models from generating visually undesirable or even harmful content. However, current removal methods heavily rely on manually crafted text prompts, making it challenging to achieve a high erasure (efficacy) while minimizing the impact on other benign concepts (usability). In this paper, we attribute the limitations to the inherent gap between the text and image modalities, which makes it hard to transfer the intricately entangled concept knowledge from text prompts to the image generation process. To address this, we propose a novel solution by directly integrating visual supervision into the erasure process, introducing the first text-image Collaborative Concept Erasing (Co-Erasing) framework. Specifically, Co-Erasing describes the concept jointly by text prompts and the corresponding undesirable images induced by the prompts, and then reduces the generating probability of the target concept through negative guidance. This approach effectively bypasses the knowledge gap between text and image, significantly enhancing erasure efficacy. Additionally, we design a text-guided image concept refinement strategy that directs the model to focus on visual features most relevant to the specified text concept, minimizing disruption to other benign concepts. Finally, comprehensive experiments suggest that Co-Erasing outperforms state-of-the-art erasure approaches significantly with a better trade-off between efficacy and usability. Codes are available at https://github.com/Ferry-Li/Co-Erasing.
Single hyperspectral image super-resolution (SHSR) aims to restore high-resolution images from low-resolution hyperspectral images. Recently, the Visual Mamba model has achieved an impressive balance between performance and computational efficiency. However, due to its 1D scanning paradigm, the model may suffer from potential artifacts during image generation. To address this issue, we propose HSRMamba. While maintaining the computational efficiency of Visual Mamba, we introduce a strip-based scanning scheme to effectively reduce artifacts from global unidirectional scanning. Additionally, HSRMamba uses wavelet decomposition to alleviate modal conflicts between high-frequency spatial features and low-frequency spectral features, further improving super-resolution performance. Extensive experiments show that HSRMamba not only excels in reducing computational load and model size but also outperforms existing methods, achieving state-of-the-art results.
While diffusion models have gained prominence in image synthesis, their generative pre-training has been shown to yield discriminative representations, paving the way towards unified visual generation and understanding. However, two key questions remain: 1) Can these representations be leveraged to improve the training of diffusion models themselves, rather than solely benefiting downstream tasks? 2) Can the feature quality be enhanced to rival or even surpass modern self-supervised learners, without compromising generative capability? This work addresses these questions by introducing self-conditioning, a straightforward yet effective mechanism that internally leverages the rich semantics inherent in denoising network to guide its own decoding layers, forming a tighter bottleneck that condenses high-level semantics to improve generation. Results are compelling: our method boosts both generation FID and recognition accuracy with 1% computational overhead and generalizes across diverse diffusion architectures. Crucially, self-conditioning facilitates an effective integration of discriminative techniques, such as contrastive self-distillation, directly into diffusion models without sacrificing generation quality. Extensive experiments on pixel-space and latent-space datasets show that in linear evaluations, our enhanced diffusion models, particularly UViT and DiT, serve as strong representation learners, surpassing various self-supervised models.
Generative modeling has emerged as a promising direction in computational pathology, offering capabilities such as data-efficient learning, synthetic data augmentation, and multimodal representation across diverse diagnostic tasks. This review provides a comprehensive synthesis of recent progress in the field, organized into four key domains: image generation, text generation, multimodal image-text generation, and other generative applications, including spatial simulation and molecular inference. By analyzing over 150 representative studies, we trace the evolution of generative architectures from early generative adversarial networks to recent advances in diffusion models and foundation models with generative capabilities. We further examine the datasets and evaluation protocols commonly used in this domain and highlight ongoing limitations, including challenges in generating high-fidelity whole slide images, clinical interpretability, and concerns related to the ethical and legal implications of synthetic data. The review concludes with a discussion of open challenges and prospective research directions, with an emphasis on developing unified, multimodal, and clinically deployable generative systems. This work aims to provide a foundational reference for researchers and practitioners developing and applying generative models in computational pathology.
Text-to-image (T2I) diffusion/flow models have drawn considerable attention recently due to their remarkable ability to deliver flexible visual creations. Still, high-resolution image synthesis presents formidable challenges due to the scarcity and complexity of high-resolution content. Recent approaches have investigated training-free strategies to enable high-resolution image synthesis with pre-trained models. However, these techniques often struggle with generating high-quality visuals and tend to exhibit artifacts or low-fidelity details, as they typically rely solely on the endpoint of the low-resolution sampling trajectory while neglecting intermediate states that are critical for preserving structure and synthesizing finer detail. To this end, we present HiFlow, a training-free and model-agnostic framework to unlock the resolution potential of pre-trained flow models. Specifically, HiFlow establishes a virtual reference flow within the high-resolution space that effectively captures the characteristics of low-resolution flow information, offering guidance for high-resolution generation through three key aspects: initialization alignment for low-frequency consistency, direction alignment for structure preservation, and acceleration alignment for detail fidelity. By leveraging such flow-aligned guidance, HiFlow substantially elevates the quality of high-resolution image synthesis of T2I models and demonstrates versatility across their personalized variants. Extensive experiments validate HiFlow's capability in achieving superior high-resolution image quality over state-of-the-art methods.
Recent advances in text-to-image diffusion models, particularly Stable Diffusion, have enabled the generation of highly detailed and semantically rich images. However, personalizing these models to represent novel subjects based on a few reference images remains challenging. This often leads to catastrophic forgetting, overfitting, or large computational overhead.We propose a two-stage pipeline that addresses these limitations by leveraging LoRA-based fine-tuning on the attention weights within the U-Net of the Stable Diffusion XL (SDXL) model. First, we use the unmodified SDXL to generate a generic scene by replacing the subject with its class label. Then, we selectively insert the personalized subject through a segmentation-driven image-to-image (Img2Img) pipeline that uses the trained LoRA weights.This framework isolates the subject encoding from the overall composition, thus preserving SDXL's broader generative capabilities while integrating the new subject in a high-fidelity manner. Our method achieves a DINO similarity score of 0.789 on SDXL, outperforming existing personalized text-to-image approaches.
Robotic imitation learning has advanced from solving static tasks to addressing dynamic interaction scenarios, but testing and evaluation remain costly and challenging due to the need for real-time interaction with dynamic environments. We propose EnerVerse-AC (EVAC), an action-conditional world model that generates future visual observations based on an agent's predicted actions, enabling realistic and controllable robotic inference. Building on prior architectures, EVAC introduces a multi-level action-conditioning mechanism and ray map encoding for dynamic multi-view image generation while expanding training data with diverse failure trajectories to improve generalization. As both a data engine and evaluator, EVAC augments human-collected trajectories into diverse datasets and generates realistic, action-conditioned video observations for policy testing, eliminating the need for physical robots or complex simulations. This approach significantly reduces costs while maintaining high fidelity in robotic manipulation evaluation. Extensive experiments validate the effectiveness of our method. Code, checkpoints, and datasets can be found at <https://annaj2178.github.io/EnerverseAC.github.io>.
Existing vision tokenization isolates the optimization of vision tokenizers from downstream training, implicitly assuming the visual tokens can generalize well across various tasks, e.g., image generation and visual question answering. The vision tokenizer optimized for low-level reconstruction is agnostic to downstream tasks requiring varied representations and semantics. This decoupled paradigm introduces a critical misalignment: The loss of the vision tokenization can be the representation bottleneck for target tasks. For example, errors in tokenizing text in a given image lead to poor results when recognizing or generating them. To address this, we propose ETT, an end-to-end vision tokenizer tuning approach that enables joint optimization between vision tokenization and target autoregressive tasks. Unlike prior autoregressive models that use only discrete indices from a frozen vision tokenizer, ETT leverages the visual embeddings of the tokenizer codebook, and optimizes the vision tokenizers end-to-end with both reconstruction and caption objectives. ETT can be seamlessly integrated into existing training pipelines with minimal architecture modifications. Our ETT is simple to implement and integrate, without the need to adjust the original codebooks or architectures of the employed large language models. Extensive experiments demonstrate that our proposed end-to-end vision tokenizer tuning unlocks significant performance gains, i.e., 2-6% for multimodal understanding and visual generation tasks compared to frozen tokenizer baselines, while preserving the original reconstruction capability. We hope this very simple and strong method can empower multimodal foundation models besides image generation and understanding.
This paper does not describe a new method; instead, it provides a thorough exploration of an important yet understudied design space related to recent advances in text-to-image synthesis -- specifically, the deep fusion of large language models (LLMs) and diffusion transformers (DiTs) for multi-modal generation. Previous studies mainly focused on overall system performance rather than detailed comparisons with alternative methods, and key design details and training recipes were often left undisclosed. These gaps create uncertainty about the real potential of this approach. To fill these gaps, we conduct an empirical study on text-to-image generation, performing controlled comparisons with established baselines, analyzing important design choices, and providing a clear, reproducible recipe for training at scale. We hope this work offers meaningful data points and practical guidelines for future research in multi-modal generation.
We present a method that reduces, by an order of magnitude, the time and memory needed to train multi-task vision-language robotic diffusion policies. This improvement arises from a previously underexplored distinction between action diffusion and the image diffusion techniques that inspired it: In image generation, the target is high-dimensional. By contrast, in action generation, the dimensionality of the target is comparatively small, and only the image condition is high-dimensional. Our approach, \emph{Mini Diffuser}, exploits this asymmetry by introducing \emph{two-level minibatching}, which pairs multiple noised action samples with each vision-language condition, instead of the conventional one-to-one sampling strategy. To support this batching scheme, we introduce architectural adaptations to the diffusion transformer that prevent information leakage across samples while maintaining full conditioning access. In RLBench simulations, Mini-Diffuser achieves 95\% of the performance of state-of-the-art multi-task diffusion policies, while using only 5\% of the training time and 7\% of the memory. Real-world experiments further validate that Mini-Diffuser preserves the key strengths of diffusion-based policies, including the ability to model multimodal action distributions and produce behavior conditioned on diverse perceptual inputs. Code available at mini-diffuse-actor.github.io
Score-based generative models (SBM), also known as diffusion models, are the de facto state of the art for image synthesis. Despite their unparalleled performance, SBMs have recently been in the spotlight for being tricked into creating not-safe-for-work (NSFW) content, such as violent images and non-consensual nudity. Current approaches that prevent unsafe generation are based on the models' own knowledge, and the majority of them require fine-tuning. This article explores the use of external sources for ensuring safe outputs in SBMs. Our safe-for-work (SFW) sampler implements a Conditional Trajectory Correction step that guides the samples away from undesired regions in the ambient space using multimodal models as the source of conditioning. Furthermore, using Contrastive Language Image Pre-training (CLIP), our method admits user-defined NSFW classes, which can vary in different settings. Our experiments on the text-to-image SBM Stable Diffusion validate that the proposed SFW sampler effectively reduces the generation of explicit content while being competitive with other fine-tuning-based approaches, as assessed via independent NSFW detectors. Moreover, we evaluate the impact of the SFW sampler on image quality and show that the proposed correction scheme comes at a minor cost with negligible effect on samples not needing correction. Our study confirms the suitability of the SFW sampler towards aligned SBM models and the potential of using model-agnostic conditioning for the prevention of unwanted images.
The field of text-to-image generation has undergone significant advancements with the introduction of diffusion models. Nevertheless, the challenge of editing real images persists, as most methods are either computationally intensive or produce poor reconstructions. This paper introduces SAGE (Self-Attention Guidance for image Editing) - a novel technique leveraging pre-trained diffusion models for image editing. SAGE builds upon the DDIM algorithm and incorporates a novel guidance mechanism utilizing the self-attention layers of the diffusion U-Net. This mechanism computes a reconstruction objective based on attention maps generated during the inverse DDIM process, enabling efficient reconstruction of unedited regions without the need to precisely reconstruct the entire input image. Thus, SAGE directly addresses the key challenges in image editing. The superiority of SAGE over other methods is demonstrated through quantitative and qualitative evaluations and confirmed by a statistically validated comprehensive user study, in which all 47 surveyed users preferred SAGE over competing methods. Additionally, SAGE ranks as the top-performing method in seven out of 10 quantitative analyses and secures second and third places in the remaining three.
Unifying image understanding and generation has gained growing attention in recent research on multimodal models. Although design choices for image understanding have been extensively studied, the optimal model architecture and training recipe for a unified framework with image generation remain underexplored. Motivated by the strong potential of autoregressive and diffusion models for high-quality generation and scalability, we conduct a comprehensive study of their use in unified multimodal settings, with emphasis on image representations, modeling objectives, and training strategies. Grounded in these investigations, we introduce a novel approach that employs a diffusion transformer to generate semantically rich CLIP image features, in contrast to conventional VAE-based representations. This design yields both higher training efficiency and improved generative quality. Furthermore, we demonstrate that a sequential pretraining strategy for unified models-first training on image understanding and subsequently on image generation-offers practical advantages by preserving image understanding capability while developing strong image generation ability. Finally, we carefully curate a high-quality instruction-tuning dataset BLIP3o-60k for image generation by prompting GPT-4o with a diverse set of captions covering various scenes, objects, human gestures, and more. Building on our innovative model design, training recipe, and datasets, we develop BLIP3-o, a suite of state-of-the-art unified multimodal models. BLIP3-o achieves superior performance across most of the popular benchmarks spanning both image understanding and generation tasks. To facilitate future research, we fully open-source our models, including code, model weights, training scripts, and pretraining and instruction tuning datasets.