Abstract:Fine-tuning open-source large-scale VDMs for the portrait video synthesis task can result in significant improvements across multiple dimensions, such as visual quality and natural facial motion dynamics. Despite their advancements, how to achieve step distillation and reduce the substantial computational overhead of large-scale VDMs remains unexplored. To fill this gap, this paper proposes Weak-to-Strong Video Distillation (W2SVD) to mitigate both the issue of insufficient training memory and the problem of training collapse observed in vanilla DMD during the training process. Specifically, we first leverage LoRA to fine-tune the fake diffusion transformer (DiT) to address the out-of-memory issue. Then, we employ the W2S distribution matching to adjust the real DiT's parameter, subtly shifting it toward the fake DiT's parameter. This adjustment is achieved by utilizing the weak weight of the low-rank branch, effectively alleviate the conundrum where the video synthesized by the few-step generator deviates from the real data distribution, leading to inaccuracies in the KL divergence approximation. Additionally, we minimize the distance between the fake data distribution and the ground truth distribution to further enhance the visual quality of the synthesized videos. As experimentally demonstrated on HunyuanVideo, W2SVD surpasses the standard Euler, LCM, DMD and even the 28-step standard sampling in FID/FVD and VBench in 1/4-step video synthesis. The project page is in https://w2svd.github.io/W2SVD/.
Abstract:Making text-to-image (T2I) generative model sample both fast and well represents a promising research direction. Previous studies have typically focused on either enhancing the visual quality of synthesized images at the expense of sampling efficiency or dramatically accelerating sampling without improving the base model's generative capacity. Moreover, nearly all inference methods have not been able to ensure stable performance simultaneously on both diffusion models (DMs) and visual autoregressive models (ARMs). In this paper, we introduce a novel plug-and-play inference paradigm, CoRe^2, which comprises three subprocesses: Collect, Reflect, and Refine. CoRe^2 first collects classifier-free guidance (CFG) trajectories, and then use collected data to train a weak model that reflects the easy-to-learn contents while reducing number of function evaluations during inference by half. Subsequently, CoRe^2 employs weak-to-strong guidance to refine the conditional output, thereby improving the model's capacity to generate high-frequency and realistic content, which is difficult for the base model to capture. To the best of our knowledge, CoRe^2 is the first to demonstrate both efficiency and effectiveness across a wide range of DMs, including SDXL, SD3.5, and FLUX, as well as ARMs like LlamaGen. It has exhibited significant performance improvements on HPD v2, Pick-of-Pic, Drawbench, GenEval, and T2I-Compbench. Furthermore, CoRe^2 can be seamlessly integrated with the state-of-the-art Z-Sampling, outperforming it by 0.3 and 0.16 on PickScore and AES, while achieving 5.64s time saving using SD3.5.Code is released at https://github.com/xie-lab-ml/CoRe/tree/main.
Abstract:We present MagicInfinite, a novel diffusion Transformer (DiT) framework that overcomes traditional portrait animation limitations, delivering high-fidelity results across diverse character types-realistic humans, full-body figures, and stylized anime characters. It supports varied facial poses, including back-facing views, and animates single or multiple characters with input masks for precise speaker designation in multi-character scenes. Our approach tackles key challenges with three innovations: (1) 3D full-attention mechanisms with a sliding window denoising strategy, enabling infinite video generation with temporal coherence and visual quality across diverse character styles; (2) a two-stage curriculum learning scheme, integrating audio for lip sync, text for expressive dynamics, and reference images for identity preservation, enabling flexible multi-modal control over long sequences; and (3) region-specific masks with adaptive loss functions to balance global textual control and local audio guidance, supporting speaker-specific animations. Efficiency is enhanced via our innovative unified step and cfg distillation techniques, achieving a 20x inference speed boost over the basemodel: generating a 10 second 540x540p video in 10 seconds or 720x720p in 30 seconds on 8 H100 GPUs, without quality loss. Evaluations on our new benchmark demonstrate MagicInfinite's superiority in audio-lip synchronization, identity preservation, and motion naturalness across diverse scenarios. It is publicly available at https://www.hedra.com/, with examples at https://magicinfinite.github.io/.
Abstract:Real-world data often contains intrinsic ambiguity that the common single-hard-label annotation paradigm ignores. Standard training using ambiguous data with these hard labels may produce overly confident models and thus leading to poor generalization. In this paper, we propose a novel framework called Quantized Label Learning (QLL) to alleviate this issue. First, we formulate QLL as learning from (very) ambiguous data with hard labels: ideally, each ambiguous instance should be associated with a ground-truth soft-label distribution describing its corresponding probabilistic weight in each class, however, this is usually not accessible; in practice, we can only observe a quantized label, i.e., a hard label sampled (quantized) from the corresponding ground-truth soft-label distribution, of each instance, which can be seen as a biased approximation of the ground-truth soft-label. Second, we propose a Class-wise Positive-Unlabeled (CPU) risk estimator that allows us to train accurate classifiers from only ambiguous data with quantized labels. Third, to simulate ambiguous datasets with quantized labels in the real world, we design a mixing-based ambiguous data generation procedure for empirical evaluation. Experiments demonstrate that our CPU method can significantly improve model generalization performance and outperform the baselines.
Abstract:Large diffusion models have become mainstream generative models in both academic studies and industrial AIGC applications. Recently, a number of works further explored how to employ the power of large diffusion models as zero-shot classifiers. While recent zero-shot diffusion-based classifiers have made performance advancement on benchmark datasets, they still suffered badly from extremely slow classification speed (e.g., ~1000 seconds per classifying single image on ImageNet). The extremely slow classification speed strongly prohibits existing zero-shot diffusion-based classifiers from practical applications. In this paper, we propose an embarrassingly simple and efficient zero-shot Gaussian Diffusion Classifiers (GDC) via pretrained text-to-image diffusion models and DINOv2. The proposed GDC can not only significantly surpass previous zero-shot diffusion-based classifiers by over 10 points (61.40% - 71.44%) on ImageNet, but also accelerate more than 30000 times (1000 - 0.03 seconds) classifying a single image on ImageNet. Additionally, it provides probability interpretation of the results. Our extensive experiments further demonstrate that GDC can achieve highly competitive zero-shot classification performance over various datasets and can promisingly self-improve with stronger diffusion models. To the best of our knowledge, the proposed GDC is the first zero-shot diffusionbased classifier that exhibits both competitive accuracy and practical efficiency.
Abstract:Diffusion models, the most popular generative paradigm so far, can inject conditional information into the generation path to guide the latent towards desired directions. However, existing text-to-image diffusion models often fail to maintain high image quality and high prompt-image alignment for those challenging prompts. To mitigate this issue and enhance existing pretrained diffusion models, we mainly made three contributions in this paper. First, we propose diffusion self-reflection that alternately performs denoising and inversion and demonstrate that such diffusion self-reflection can leverage the guidance gap between denoising and inversion to capture prompt-related semantic information with theoretical and empirical evidence. Second, motivated by theoretical analysis, we derive Zigzag Diffusion Sampling (Z-Sampling), a novel self-reflection-based diffusion sampling method that leverages the guidance gap between denosing and inversion to accumulate semantic information step by step along the sampling path, leading to improved sampling results. Moreover, as a plug-and-play method, Z-Sampling can be generally applied to various diffusion models (e.g., accelerated ones and Transformer-based ones) with very limited coding and computational costs. Third, our extensive experiments demonstrate that Z-Sampling can generally and significantly enhance generation quality across various benchmark datasets, diffusion models, and performance evaluation metrics. For example, DreamShaper with Z-Sampling can self-improve with the HPSv2 winning rate up to 94% over the original results. Moreover, Z-Sampling can further enhance existing diffusion models combined with other orthogonal methods, including Diffusion-DPO.
Abstract:Text-to-image diffusion models (DMs) develop at an unprecedented pace, supported by thorough theoretical exploration and empirical analysis. Unfortunately, the discrepancy between DMs and autoregressive models (ARMs) complicates the path toward achieving the goal of unified vision and language generation. Recently, the masked generative Transformer (MGT) serves as a promising intermediary between DM and ARM by predicting randomly masked image tokens (i.e., masked image modeling), combining the efficiency of DM with the discrete token nature of ARM. However, we find that the comprehensive analyses regarding the inference for MGT are virtually non-existent, and thus we aim to present positive design choices to fill this gap. We modify and re-design a set of DM-based inference techniques for MGT and further elucidate their performance on MGT. We also discuss the approach to correcting token's distribution to enhance inference. Extensive experiments and empirical analyses lead to concrete and effective design choices, and these design choices can be merged to achieve further performance gains. For instance, in terms of enhanced inference, we achieve winning rates of approximately 70% compared to vanilla sampling on HPS v2 with the recent SOTA MGT Meissonic. Our contributions have the potential to further enhance the capabilities and future development of MGTs.
Abstract:Text-to-image diffusion model is a popular paradigm that synthesizes personalized images by providing a text prompt and a random Gaussian noise. While people observe that some noises are ``golden noises'' that can achieve better text-image alignment and higher human preference than others, we still lack a machine learning framework to obtain those golden noises. To learn golden noises for diffusion sampling, we mainly make three contributions in this paper. First, we identify a new concept termed the \textit{noise prompt}, which aims at turning a random Gaussian noise into a golden noise by adding a small desirable perturbation derived from the text prompt. Following the concept, we first formulate the \textit{noise prompt learning} framework that systematically learns ``prompted'' golden noise associated with a text prompt for diffusion models. Second, we design a noise prompt data collection pipeline and collect a large-scale \textit{noise prompt dataset}~(NPD) that contains 100k pairs of random noises and golden noises with the associated text prompts. With the prepared NPD as the training dataset, we trained a small \textit{noise prompt network}~(NPNet) that can directly learn to transform a random noise into a golden noise. The learned golden noise perturbation can be considered as a kind of prompt for noise, as it is rich in semantic information and tailored to the given text prompt. Third, our extensive experiments demonstrate the impressive effectiveness and generalization of NPNet on improving the quality of synthesized images across various diffusion models, including SDXL, DreamShaper-xl-v2-turbo, and Hunyuan-DiT. Moreover, NPNet is a small and efficient controller that acts as a plug-and-play module with very limited additional inference and computational costs, as it just provides a golden noise instead of a random noise without accessing the original pipeline.
Abstract:Recent advancements in computational chemistry have leveraged the power of trans-former-based language models, such as MoLFormer, pre-trained using a vast amount of simplified molecular-input line-entry system (SMILES) sequences, to understand and predict molecular properties and activities, a critical step in fields like drug discovery and materials science. To further improve performance, researchers have introduced graph neural networks with graph-based molecular representations, such as GEM, incorporating the topology, geometry, 2D or even 3D structures of molecules into pre-training. While most of molecular graphs in existing studies were automatically converted from SMILES sequences, it is to assume that transformer-based language models might be able to implicitly learn structure-aware representations from SMILES sequences. In this paper, we propose \ours{} -- a SMILES-based \underline{\em M}olecular \underline{\em L}anguage \underline{\em M}odel, which randomly masking SMILES subsequences corresponding to specific molecular \underline{\em F}unctional \underline{\em G}roups to incorporate structure information of atoms during the pre-training phase. This technique aims to compel the model to better infer molecular structures and properties, thus enhancing its predictive capabilities. Extensive experimental evaluations across 11 benchmark classification and regression tasks in the chemical domain demonstrate the robustness and superiority of \ours{}. Our findings reveal that \ours{} outperforms existing pre-training models, either based on SMILES or graphs, in 9 out of the 11 downstream tasks, ranking as a close second in the remaining ones.
Abstract:The multi-step sampling mechanism, a key feature of visual diffusion models, has significant potential to replicate the success of OpenAI's Strawberry in enhancing performance by increasing the inference computational cost. Sufficient prior studies have demonstrated that correctly scaling up computation in the sampling process can successfully lead to improved generation quality, enhanced image editing, and compositional generalization. While there have been rapid advancements in developing inference-heavy algorithms for improved image generation, relatively little work has explored inference scaling laws in video diffusion models (VDMs). Furthermore, existing research shows only minimal performance gains that are perceptible to the naked eye. To address this, we design a novel training-free algorithm IV-Mixed Sampler that leverages the strengths of image diffusion models (IDMs) to assist VDMs surpass their current capabilities. The core of IV-Mixed Sampler is to use IDMs to significantly enhance the quality of each video frame and VDMs ensure the temporal coherence of the video during the sampling process. Our experiments have demonstrated that IV-Mixed Sampler achieves state-of-the-art performance on 4 benchmarks including UCF-101-FVD, MSR-VTT-FVD, Chronomagic-Bench-150, and Chronomagic-Bench-1649. For example, the open-source Animatediff with IV-Mixed Sampler reduces the UMT-FVD score from 275.2 to 228.6, closing to 223.1 from the closed-source Pika-2.0.