Diffusion models have been widely adopted in image and video generation. However, their complex network architecture leads to high inference overhead for its generation process. Existing diffusion quantization methods primarily focus on the quantization of the model structure while ignoring the impact of time-steps variation during sampling. At the same time, most current approaches fail to account for significant activations that cannot be eliminated, resulting in substantial performance degradation after quantization. To address these issues, we propose Time-Rotation Diffusion Quantization (TR-DQ), a novel quantization method incorporating time-step and rotation-based optimization. TR-DQ first divides the sampling process based on time-steps and applies a rotation matrix to smooth activations and weights dynamically. For different time-steps, a dedicated hyperparameter is introduced for adaptive timing modeling, which enables dynamic quantization across different time steps. Additionally, we also explore the compression potential of Classifier-Free Guidance (CFG-wise) to establish a foundation for subsequent work. TR-DQ achieves state-of-the-art (SOTA) performance on image generation and video generation tasks and a 1.38-1.89x speedup and 1.97-2.58x memory reduction in inference compared to existing quantization methods.
Generating coherent long-form video sequences from discrete input using only text prompts is a critical task in content creation. While diffusion-based models excel at short video synthesis, long-form storytelling from text remains largely unexplored and a challenge due to challenges pertaining to temporal coherency, preserving semantic meaning and action continuity across the video. We introduce a novel storytelling approach to enable seamless video generation with natural action transitions and structured narratives. We present a bidirectional time-weighted latent blending strategy to ensure temporal consistency between segments of the long-form video being generated. Further, our method extends the Black-Scholes algorithm from prompt mixing for image generation to video generation, enabling controlled motion evolution through structured text conditioning. To further enhance motion continuity, we propose a semantic action representation framework to encode high-level action semantics into the blending process, dynamically adjusting transitions based on action similarity, ensuring smooth yet adaptable motion changes. Latent space blending maintains spatial coherence between objects in a scene, while time-weighted blending enforces bidirectional constraints for temporal consistency. This integrative approach prevents abrupt transitions while ensuring fluid storytelling. Extensive experiments demonstrate significant improvements over baselines, achieving temporally consistent and visually compelling video narratives without any additional training. Our approach bridges the gap between short clips and extended video to establish a new paradigm in GenAI-driven video synthesis from text.
We propose V2Flow, a novel tokenizer that produces discrete visual tokens capable of high-fidelity reconstruction, while ensuring structural and latent distribution alignment with the vocabulary space of large language models (LLMs). Leveraging this tight visual-vocabulary coupling, V2Flow enables autoregressive visual generation on top of existing LLMs. Our approach formulates visual tokenization as a flow-matching problem, aiming to learn a mapping from a standard normal prior to the continuous image distribution, conditioned on token sequences embedded within the LLMs vocabulary space. The effectiveness of V2Flow stems from two core designs. First, we propose a Visual Vocabulary resampler, which compresses visual data into compact token sequences, with each represented as a soft categorical distribution over LLM's vocabulary. This allows seamless integration of visual tokens into existing LLMs for autoregressive visual generation. Second, we present a masked autoregressive Rectified-Flow decoder, employing a masked transformer encoder-decoder to refine visual tokens into contextually enriched embeddings. These embeddings then condition a dedicated velocity field for precise reconstruction. Additionally, an autoregressive rectified-flow sampling strategy is incorporated, ensuring flexible sequence lengths while preserving competitive reconstruction quality. Extensive experiments show that V2Flow outperforms mainstream VQ-based tokenizers and facilitates autoregressive visual generation on top of existing. https://github.com/zhangguiwei610/V2Flow
Recent advances in text-to-image diffusion models enable photorealistic image generation, but they also risk producing malicious content, such as NSFW images. To mitigate risk, concept erasure methods are studied to facilitate the model to unlearn specific concepts. However, current studies struggle to fully erase malicious concepts implicitly embedded in prompts (e.g., metaphorical expressions or adversarial prompts) while preserving the model's normal generation capability. To address this challenge, our study proposes TRCE, using a two-stage concept erasure strategy to achieve an effective trade-off between reliable erasure and knowledge preservation. Firstly, TRCE starts by erasing the malicious semantics implicitly embedded in textual prompts. By identifying a critical mapping objective(i.e., the [EoT] embedding), we optimize the cross-attention layers to map malicious prompts to contextually similar prompts but with safe concepts. This step prevents the model from being overly influenced by malicious semantics during the denoising process. Following this, considering the deterministic properties of the sampling trajectory of the diffusion model, TRCE further steers the early denoising prediction toward the safe direction and away from the unsafe one through contrastive learning, thus further avoiding the generation of malicious content. Finally, we conduct comprehensive evaluations of TRCE on multiple malicious concept erasure benchmarks, and the results demonstrate its effectiveness in erasing malicious concepts while better preserving the model's original generation ability. The code is available at: http://github.com/ddgoodgood/TRCE. CAUTION: This paper includes model-generated content that may contain offensive material.
Autoregressive models have achieved promising results in natural language processing. However, for image generation tasks, they encounter substantial challenges in effectively capturing long-range dependencies, managing computational costs, and most crucially, defining meaningful autoregressive sequences that reflect natural image hierarchies. To address these issues, we present \textbf{N}ext-\textbf{F}requency \textbf{I}mage \textbf{G}eneration (\textbf{NFIG}), a novel framework that decomposes the image generation process into multiple frequency-guided stages. Our approach first generates low-frequency components to establish global structure with fewer tokens, then progressively adds higher-frequency details, following the natural spectral hierarchy of images. This principled autoregressive sequence not only improves the quality of generated images by better capturing true causal relationships between image components, but also significantly reduces computational overhead during inference. Extensive experiments demonstrate that NFIG achieves state-of-the-art performance with fewer steps, offering a more efficient solution for image generation, with 1.25$\times$ speedup compared to VAR-d20 while achieving better performance (FID: 2.81) on the ImageNet-256 benchmark. We hope that our insight of incorporating frequency-domain knowledge to guide autoregressive sequence design will shed light on future research. We will make our code publicly available upon acceptance of the paper.
The inherent difficulty in acquiring accurately co-registered RGB-hyperspectral image (HSI) pairs has significantly impeded the practical deployment of current data-driven Hyperspectral Image Generation (HIG) networks in engineering applications. Gleichzeitig, the ill-posed nature of the aligning constraints, compounded with the complexities of mining cross-domain features, also hinders the advancement of unpaired HIG (UnHIG) tasks. In this paper, we conquer these challenges by modeling the UnHIG to range space interaction and compensations of null space through Range-Null Space Decomposition (RND) methodology. Specifically, the introduced contrastive learning effectively aligns the geometric and spectral distributions of unpaired data by building the interaction of range space, considering the consistent feature in degradation process. Following this, we map the frequency representations of dual-domain input and thoroughly mining the null space, like degraded and high-frequency components, through the proposed Non-uniform Kolmogorov-Arnold Networks. Extensive comparative experiments demonstrate that it establishes a new benchmark in UnHIG.
Personalized image generation with text-to-image diffusion models generates unseen images based on reference image content. Zero-shot adapter methods such as IP-Adapter and OminiControl are especially interesting because they do not require test-time fine-tuning. However, they struggle to balance preserving personalized content and adherence to the text prompt. We identify a critical design flaw resulting in this performance gap: current adapters inadequately integrate personalization images with the textual descriptions. The generated images, therefore, replicate the personalized content rather than adhere to the text prompt instructions. Yet the base text-to-image has strong conceptual understanding capabilities that can be leveraged. We propose Conceptrol, a simple yet effective framework that enhances zero-shot adapters without adding computational overhead. Conceptrol constrains the attention of visual specification with a textual concept mask that improves subject-driven generation capabilities. It achieves as much as 89% improvement on personalization benchmarks over the vanilla IP-Adapter and can even outperform fine-tuning approaches such as Dreambooth LoRA. The source code is available at https://github.com/QY-H00/Conceptrol.
Unified models (UniMs) for multimodal understanding and generation have recently received much attention in the area of vision and language. Existing UniMs are designed to simultaneously learn both multimodal understanding and generation capabilities, demanding substantial computational resources, and often struggle to generate interleaved text-image. We present ARMOR, a resource-efficient and pure autoregressive framework that achieves both understanding and generation by fine-tuning existing multimodal large language models (MLLMs). Specifically, ARMOR extends existing MLLMs from three perspectives: (1) For model architecture, an asymmetric encoder-decoder architecture with a forward-switching mechanism is introduced to unify embedding space integrating textual and visual modalities for enabling natural text-image interleaved generation with minimal computational overhead. (2) For training data, a meticulously curated, high-quality interleaved dataset is collected for fine-tuning MLLMs. (3) For the training algorithm, we propose a ``what or how to generate" algorithm to empower existing MLLMs with multimodal generation capabilities while preserving their multimodal understanding capabilities, through three progressive training stages based on the collected dataset. Experimental results demonstrate that ARMOR upgrades existing MLLMs to UniMs with promising image generation capabilities, using limited training resources. Our code will be released soon at https://armor.github.io.
Recent advancements in text-to-image generation have spurred interest in personalized human image generation, which aims to create novel images featuring specific human identities as reference images indicate. Although existing methods achieve high-fidelity identity preservation, they often struggle with limited multi-ID usability and inadequate facial editability. We present DynamicID, a tuning-free framework supported by a dual-stage training paradigm that inherently facilitates both single-ID and multi-ID personalized generation with high fidelity and flexible facial editability. Our key innovations include: 1) Semantic-Activated Attention (SAA), which employs query-level activation gating to minimize disruption to the original model when injecting ID features and achieve multi-ID personalization without requiring multi-ID samples during training. 2) Identity-Motion Reconfigurator (IMR), which leverages contrastive learning to effectively disentangle and re-entangle facial motion and identity features, thereby enabling flexible facial editing. Additionally, we have developed a curated VariFace-10k facial dataset, comprising 10k unique individuals, each represented by 35 distinct facial images. Experimental results demonstrate that DynamicID outperforms state-of-the-art methods in identity fidelity, facial editability, and multi-ID personalization capability.
Optical illusion hidden picture is an interesting visual perceptual phenomenon where an image is cleverly integrated into another picture in a way that is not immediately obvious to the viewer. Established on the off-the-shelf text-to-image (T2I) diffusion model, we propose a novel training-free text-guided image-to-image (I2I) translation framework dubbed as \textbf{P}hase-\textbf{T}ransferred \textbf{Diffusion} Model (PTDiffusion) for hidden art syntheses. PTDiffusion harmoniously embeds an input reference image into arbitrary scenes described by the text prompts, producing illusion images exhibiting hidden visual cues of the reference image. At the heart of our method is a plug-and-play phase transfer mechanism that dynamically and progressively transplants diffusion features' phase spectrum from the denoising process to reconstruct the reference image into the one to sample the generated illusion image, realizing deep fusion of the reference structural information and the textual semantic information in the diffusion model latent space. Furthermore, we propose asynchronous phase transfer to enable flexible control to the degree of hidden content discernability. Our method bypasses any model training and fine-tuning process, all while substantially outperforming related text-guided I2I methods in image generation quality, text fidelity, visual discernibility, and contextual naturalness for illusion picture synthesis, as demonstrated by extensive qualitative and quantitative experiments. Our project is publically available at \href{https://xianggao1102.github.io/PTDiffusion_webpage/}{this web page}.
Faces synthesized by diffusion models (DMs) with high-quality and controllable attributes pose a significant challenge for Deepfake detection. Most state-of-the-art detectors only yield a binary decision, incapable of forgery localization, attribution of forgery methods, and providing analysis on the cause of forgeries. In this work, we integrate Multimodal Large Language Models (MLLMs) within DM-based face forensics, and propose a fine-grained analysis triad framework called VLForgery, that can 1) predict falsified facial images; 2) locate the falsified face regions subjected to partial synthesis; and 3) attribute the synthesis with specific generators. To achieve the above goals, we introduce VLF (Visual Language Forensics), a novel and diverse synthesis face dataset designed to facilitate rich interactions between Visual and Language modalities in MLLMs. Additionally, we propose an extrinsic knowledge-guided description method, termed EkCot, which leverages knowledge from the image generation pipeline to enable MLLMs to quickly capture image content. Furthermore, we introduce a low-level vision comparison pipeline designed to identify differential features between real and fake that MLLMs can inherently understand. These features are then incorporated into EkCot, enhancing its ability to analyze forgeries in a structured manner, following the sequence of detection, localization, and attribution. Extensive experiments demonstrate that VLForgery outperforms other state-of-the-art forensic approaches in detection accuracy, with additional potential for falsified region localization and attribution analysis.
Text-to-image (T2I) models are well known for their ability to produce highly realistic images, while multimodal large language models (MLLMs) are renowned for their proficiency in understanding and integrating multiple modalities. However, currently there is no straightforward and efficient framework to transfer the multimodal comprehension abilities of MLLMs to T2I models to enable them to understand multimodal inputs. In this paper, we propose the X2I framework, which endows Diffusion Transformer (DiT) models with the capability to comprehend various modalities, including multilingual text, screenshot documents, images, videos, and audio. X2I is trained using merely 100K English corpus with 160 GPU hours. Building on the DiT teacher model, we adopt an innovative distillation method to extract the inference capabilities of the teacher model and design a lightweight AlignNet structure to serve as an intermediate bridge. Compared to the teacher model, X2I shows a decrease in performance degradation of less than 1\% while gaining various multimodal understanding abilities, including multilingual to image, image to image, image-text to image, video to image, audio to image, and utilizing creative fusion to enhance imagery. Furthermore, it is applicable for LoRA training in the context of image-text to image generation, filling a void in the industry in this area. We further design a simple LightControl to enhance the fidelity of instructional image editing. Finally, extensive experiments demonstrate the effectiveness, efficiency, multifunctionality, and transferability of our X2I. The open-source code and checkpoints for X2I can be found at the following link: https://github.com/OPPO-Mente-Lab/X2I.
Recent studies have highlighted the interplay between diffusion models and representation learning. Intermediate representations from diffusion models can be leveraged for downstream visual tasks, while self-supervised vision models can enhance the convergence and generation quality of diffusion models. However, transferring pretrained weights from vision models to diffusion models is challenging due to input mismatches and the use of latent spaces. To address these challenges, we propose Unified Self-supervised Pretraining (USP), a framework that initializes diffusion models via masked latent modeling in a Variational Autoencoder (VAE) latent space. USP achieves comparable performance in understanding tasks while significantly improving the convergence speed and generation quality of diffusion models. Our code will be publicly available at https://github.com/cxxgtxy/USP.
Collecting and annotating medical images is a time-consuming and resource-intensive task. However, generating synthetic data through models such as Diffusion offers a cost-effective alternative. This paper introduces a new method for the automatic generation of accurate semantic masks from synthetic lung X-ray images based on a stable diffusion model trained on text-image pairs. This method uses cross-attention mapping between text and image to extend text-driven image synthesis to semantic mask generation. It employs text-guided cross-attention information to identify specific areas in an image and combines this with innovative techniques to produce high-resolution, class-differentiated pixel masks. This approach significantly reduces the costs associated with data collection and annotation. The experimental results demonstrate that segmentation models trained on synthetic data generated using the method are comparable to, and in some cases even better than, models trained on real datasets. This shows the effectiveness of the method and its potential to revolutionize medical image analysis.
Generative modeling is widely regarded as one of the most essential problems in today's AI community, with text-to-image generation having gained unprecedented real-world impacts. Among various approaches, diffusion models have achieved remarkable success and have become the de facto solution for text-to-image generation. However, despite their impressive performance, these models exhibit fundamental limitations in adhering to numerical constraints in user instructions, frequently generating images with an incorrect number of objects. While several prior works have mentioned this issue, a comprehensive and rigorous evaluation of this limitation remains lacking. To address this gap, we introduce T2ICountBench, a novel benchmark designed to rigorously evaluate the counting ability of state-of-the-art text-to-image diffusion models. Our benchmark encompasses a diverse set of generative models, including both open-source and private systems. It explicitly isolates counting performance from other capabilities, provides structured difficulty levels, and incorporates human evaluations to ensure high reliability. Extensive evaluations with T2ICountBench reveal that all state-of-the-art diffusion models fail to generate the correct number of objects, with accuracy dropping significantly as the number of objects increases. Additionally, an exploratory study on prompt refinement demonstrates that such simple interventions generally do not improve counting accuracy. Our findings highlight the inherent challenges in numerical understanding within diffusion models and point to promising directions for future improvements.
Text-to-image generative models have made significant advancements in recent years; however, accurately capturing intricate details in textual prompts, such as entity missing, attribute binding errors, and incorrect relationships remains a formidable challenge. In response, we present an innovative, training-free method that directly addresses these challenges by incorporating tailored objectives to account for textual constraints. Unlike layout-based approaches that enforce rigid structures and limit diversity, our proposed approach offers a more flexible arrangement of the scene by imposing just the extracted constraints from the text, without any unnecessary additions. These constraints are formulated as losses-entity missing, entity mixing, attribute binding, and spatial relationships, integrated into a unified loss that is applied in the first generation stage. Furthermore, we introduce a feedback-driven system for fine-grained initial noise refinement. This system integrates a verifier that evaluates the generated image, identifies inconsistencies, and provides corrective feedback. Leveraging this feedback, our refinement method first targets the unmet constraints by refining the faulty attention maps caused by initial noise, through the optimization of selective losses associated with these constraints. Subsequently, our unified loss function is reapplied to proceed the second generation phase. Experimental results demonstrate that our method, relying solely on our proposed objective functions, significantly enhances compositionality, achieving a 24% improvement in human evaluation and a 25% gain in spatial relationships. Furthermore, our fine-grained noise refinement proves effective, boosting performance by up to 5%. Code is available at https://github.com/hadi-hosseini/noise-refinement.
Score-based diffusion models generate samples from an unknown target distribution using a time-reversed diffusion process. While such models represent state-of-the-art approaches in industrial applications such as artificial image generation, it has recently been noted that their performance can be further improved by considering injection noise with heavy tailed characteristics. Here, I present a generalization of generative diffusion processes to a wide class of non-Gaussian noise processes. I consider forward processes driven by standard Gaussian noise with super-imposed Poisson jumps representing a finite activity Levy process. The generative process is shown to be governed by a generalized score function that depends on the jump amplitude distribution. Both probability flow ODE and SDE formulations are derived using basic technical effort, and are implemented for jump amplitudes drawn from a multivariate Laplace distribution. Remarkably, for the problem of capturing a heavy-tailed target distribution, the jump-diffusion Laplace model outperforms models driven by alpha-stable noise despite not containing any heavy-tailed characteristics. The framework can be readily applied to other jump statistics that could further improve on the performance of standard diffusion models.
Text-to-Image (T2I) models are capable of generating high-quality artistic creations and visual content. However, existing research and evaluation standards predominantly focus on image realism and shallow text-image alignment, lacking a comprehensive assessment of complex semantic understanding and world knowledge integration in text to image generation. To address this challenge, we propose $\textbf{WISE}$, the first benchmark specifically designed for $\textbf{W}$orld Knowledge-$\textbf{I}$nformed $\textbf{S}$emantic $\textbf{E}$valuation. WISE moves beyond simple word-pixel mapping by challenging models with 1000 meticulously crafted prompts across 25 sub-domains in cultural common sense, spatio-temporal reasoning, and natural science. To overcome the limitations of traditional CLIP metric, we introduce $\textbf{WiScore}$, a novel quantitative metric for assessing knowledge-image alignment. Through comprehensive testing of 20 models (10 dedicated T2I models and 10 unified multimodal models) using 1,000 structured prompts spanning 25 subdomains, our findings reveal significant limitations in their ability to effectively integrate and apply world knowledge during image generation, highlighting critical pathways for enhancing knowledge incorporation and application in next-generation T2I models. Code and data are available at https://github.com/PKU-YuanGroup/WISE.
Reconstructing three-dimensional (3D) structures from two-dimensional (2D) X-ray images is a valuable and efficient technique in medical applications that requires less radiation exposure than computed tomography scans. Recent approaches that use implicit neural representations have enabled the synthesis of novel views from sparse X-ray images. However, although image synthesis has improved the accuracy, the accuracy of surface shape estimation remains insufficient. Therefore, we propose a novel approach for reconstructing 3D scenes using a Neural Attenuation Surface (NeAS) that simultaneously captures the surface geometry and attenuation coefficient fields. NeAS incorporates a signed distance function (SDF), which defines the attenuation field and aids in extracting the 3D surface within the scene. We conducted experiments using simulated and authentic X-ray images, and the results demonstrated that NeAS could accurately extract 3D surfaces within a scene using only 2D X-ray images.
We investigate methods to reduce inference time and memory footprint in stable diffusion models by introducing lightweight decoders for both image and video synthesis. Traditional latent diffusion pipelines rely on large Variational Autoencoder decoders that can slow down generation and consume considerable GPU memory. We propose custom-trained decoders using lightweight Vision Transformer and Taming Transformer architectures. Experiments show up to 15% overall speed-ups for image generation on COCO2017 and up to 20 times faster decoding in the sub-module, with additional gains on UCF-101 for video tasks. Memory requirements are moderately reduced, and while there is a small drop in perceptual quality compared to the default decoder, the improvements in speed and scalability are crucial for large-scale inference scenarios such as generating 100K images. Our work is further contextualized by advances in efficient video generation, including dual masking strategies, illustrating a broader effort to improve the scalability and efficiency of generative models.