Text-to-image diffusion models have demonstrated significant capabilities to generate diverse and detailed visuals in various domains, and story visualization is emerging as a particularly promising application. However, as their use in real-world creative domains increases, the need for providing enhanced control, refinement, and the ability to modify images post-generation in a consistent manner becomes an important challenge. Existing methods often lack the flexibility to apply fine or coarse edits while maintaining visual and narrative consistency across multiple frames, preventing creators from seamlessly crafting and refining their visual stories. To address these challenges, we introduce Plot'n Polish, a zero-shot framework that enables consistent story generation and provides fine-grained control over story visualizations at various levels of detail.

Savya Khosla, Aditi Tiwari, Kushal Kafle, Simon Jenni, Handong Zhao, John Collomosse, Jing Shi. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

The advent of large Vision-Language Models (VLMs) has significantly advanced multimodal tasks, enabling more sophisticated and accurate reasoning across various applications, including image and video captioning, visual question answering, and cross-modal retrieval. Despite their superior capabilities, VLMs struggle with fine-grained image regional composition information perception. Specifically, they have difficulty accurately aligning the segmentation masks with the corresponding semantics and precisely describing the compositional aspects of the referred regions. However, compositionality – the ability to understand and generate novel combinations of known visual and textual components – is critical for facilitating coherent reasoning and understanding across modalities by VLMs. To address this issue, we propose FineCaption, a novel VLM that can recognize arbitrary masks as referential inputs and process high-resolution images for compositional image captioning at different granularity levels. To support this endeavor, we introduce CompositionCap, a new dataset for multi-grained region compositional image captioning, which introduces the task of compositional attribute-aware regional image captioning. Empirical results demonstrate the effectiveness of our proposed model compared to other state-of-the-art VLMs. Additionally, we analyze the capabilities of current VLMs in recognizing various visual prompts for compositional region image captioning, highlighting areas for improvement in VLM design and training. https://hanghuacs.github.io/FineCaption/

We introduce Visual Persona, a foundation model for text-to-image full-body human customization that, given a single in-the-wild human image, generates diverse images of the individual guided by text descriptions. Unlike prior methods that focus solely on preserving facial identity, our approach captures detailed full-body appearance, aligning with text descriptions for body structure and scene variations. Training this model requires large-scale paired human data, consisting of multiple images per individual with consistent full-body identities, which is notoriously difficult to obtain. To address this, we propose a data curation pipeline leveraging vision-language models to evaluate full-body appearance consistency, resulting in Visual Persona-500K—a dataset of 580k paired human images across 100k unique identities. For precise appearance transfer, we introduce a transformer encoder-decoder architecture adapted to a pre-trained text-to-image diffusion model, which augments the input image into distinct body regions, encodes these regions as local appearance features, and projects them into dense identity embeddings independently to condition the diffusion model for synthesizing customized images. Visual Persona consistently surpasses existing approaches, generating high-quality, customized images from in-the-wild inputs. Extensive ablation studies validate design choices, and we demonstrate the versatility of Visual Persona across various downstream tasks.

Large Multimodal Models (e.g., GPT-4, Gemini, Chameleon) have evolved into powerful tools with millions of users. However, they remain generic models and lack personalized knowledge of specific user concepts. Previous work has explored personalization for text generation, yet it remains unclear how these methods can be adapted to new modalities, such as image generation. In this paper, we introduce Yo’Chameleon, the first attempt to study personalization for large multimodal models. Given 3-5 images of a particular concept, Yo’Chameleon leverages soft-prompt tuning to embed subject-specific information to (i) answer questions about the subject and (ii) recreate pixel-level details to produce images of the subject in new contexts. Yo’Chameleon is trained with (i) a self-prompting optimization mechanism to balance performance across multiple modalities, and (ii) a "soft-positive" image generation approach to enhance image quality in a few-shot setting. Our qualitative and quantitative analyses reveal that Yo’Chameleon can learn concepts more efficiently using fewer tokens and effectively encode visual attributes, outperforming prompting baselines.

"While editing directly from life, photographers have found it too difficult to see simultaneously both the blue and the sky."John Szarkowski, William Eggleston’s Guide1Photographer and curator, Szarkowski insightfully revealed one of the notable gaps between general and aesthetic visual understanding: while the former focuses on identifying the factual element in an image (sky), the latter transcends such object identification, viewing it instead as an aesthetic component—a pure color block (blue). Such fundamental distinctions between general (detection, localization, etc.) and aesthetic (color, lighting, composition, etc.) visual understanding present a significant challenge for Multimodal Large Language Models (MLLMs). Although some recent works have made initial explorations, they are often limited to general and basic aesthetic commonsense. As a result, they frequently fall short in real-world scenarios (Fig. 1), which require extensive expertise—including photographic techniques, photo pre/post-processing knowledge, and more, to provide a detailed analysis and description. To fundamentally enhance the aesthetics understanding of MLLMs, we first introduce a novel dataset, PhotoCritique, derived from extensive discussions among professional photographers and enthusiasts, and characterized by the large scale, expertise, and diversity. Then, to better learn visual aesthetics from PhotoCritique, we furthur propose a novel model, PhotoEye, featuring a language-guided multi-view vision fusion mechanism to understand image aesthetics from multiple perspectives. Finally, we present a novel benchmark, PhotoBench, a comprehensive and professional benchmark for aesthetic visual understanding. On existing benchmarks and PhotoBench, our model demonstrates clear advantages over existing models.

Traditional alignment methods for Large Vision and Language Models (LVLMs) primarily rely on human-curated preference data. Human-generated preference data is costly; machine-generated preference data is limited in quality; and self-supervised preference data often introduces hallucinations. To overcome these limitations, we propose a novel Panel-of-Peers learning framework inspired by collaborative learning among humans. This approach leverages a panel of LVLMs, each evaluating and learning from their collective outputs through an iterative self-improvement process. By simulating a peer review system, our models generate, assess, and refine outputs in response to a curated set of prompts, mimicking a classroom learning environment. We demonstrate that this methodology enhances model performance without requiring extensive human-labeled datasets. Our experiments show significant improvement across multiple benchmarks, demonstrating the potential of peer evaluations as a scalable alternative to self-supervised alignment. Notably, we show that Panel-of-Peers increases the average score on fifteen benchmarks from 48% to 57%

Multimodal large language models (MLLMs) excel at generating highly detailed captions but often produce hallucinations. Our analysis reveals that existing hallucination detection methods struggle with detailed captions. We attribute this to the increasing reliance of MLLMs on their generated text, rather than the input image, as the sequence length grows. To address this issue, we propose a multiagent approach that leverages LLM-MLLM collaboration to correct given captions. Additionally, we introduce an evaluation framework and a benchmark dataset to facilitate the systematic analysis of detailed captions. Our experiments demonstrate that the proposed evaluation method aligns better with human judgments of factuality than existing metrics. Moreover, we show that current approaches for enhancing MLLM factuality often fail in hyper-detailed image captioning tasks. In contrast, our approach significantly enhances the factual accuracy of captions, even improving those generated by GPT-4V. Finally, we highlight a limitation of VQA-centric benchmarking by demonstrating that an MLLM’s performance on VQA benchmarks may not correlate with its ability to generate detailed image captions.

Recent Diffusion Transformers (DiTs) have shown impressive capabilities in generating single-modality content, including images, videos, and audio. However, the potential of DiTs to enable superb multimodal content creation remains underexplored. To bridge this gap, we introduce AV-DiT, a novel and efficient audio-visual diffusion transformer designed to generate high-quality, realistic videos with synchronized audio tracks. To minimize model complexity and computational costs, our AV-DiT utilizes a modality-shared DiT backbone pre-trained on image-only data, with only newly inserted adapters being trainable. This shared backbone facilitates the generation of both audio and video. Specifically, the video branch incorporates a trainable temporal attention layer into a pre-trained DiT block for capturing the temporal consistency for video generation. In addition, a small number of trainable parameters adapt the image-based DiT block to learn the acoustic characteristics for audio generation. An extra shared self-attention block reused from the DiT block, equipped with lightweight parameters, facilitates feature interaction between audio and visual modalities for alignment. Extensive experiments on the datasets demonstrate that our AV-DiT achieves state-of-the-art performance in joint audio-visual generation with significantly fewer tunable parameters. Furthermore, our results highlight that a single shared image generative backbone with modality-specific adaptations is sufficient for constructing a joint audio-video generator.

We present VIXEN - a technique that succinctly summarizes in text the visual differences between a pair of images in order to highlight any content manipulation present. Our proposed network linearly maps image features in a pairwise manner, constructing a soft prompt for a pretrained large language model. We address the challenge of low volume of training data and lack of manipulation variety in existing image difference captioning (IDC) datasets by training on synthetically manipulated images from the recent InstructPix2Pix dataset generated via prompt-to-prompt editing framework. We augment this dataset with change summaries produced via GPT-3. We show that VIXEN produces state-of-the-art, comprehensible difference captions for diverse image contents and edit types, offering a potential mitigation against misinformation disseminated via manipulated image content. Code and data are available at http://github.com/alexblck/vixen

Recently, large pretrained models (e.g., BERT, Style-GAN, CLIP) show great knowledge transfer and generalization capability on various downstream tasks within their domains. Inspired by these efforts, in this paper we propose a unified model for open-domain image editing focusing on color and tone adjustment of open-domain images while keeping their original content and structure. Our model learns a unified editing space that is more semantic, intu-itive, and easy to manipulate than the operation space (e.g., contrast, brightness, color curve) used in many existing photo editing softwares. Our model belongs to the image-to-image translation framework which consists of an image encoder and decoder, and is trained on pairs of before-and-after edited images to produce multimodal outputs. We show that by inverting image pairs into latent codes of the learned editing space, our model can be leveraged for vari-ous downstream editing tasks such as language-guided image editing, personalized editing, editing-style clustering, retrieval, etc. We extensively study the unique properties of the editing space in experiments and demonstrate superior performance on the aforementioned tasks11Code and supplementary material can be found at the project page https://jshi31.github.io/SpaceEdit.

Recently, language-guided global image editing draws increasing attention with growing application potentials. However, previous GAN-based methods are not only confined to domain-specific, low-resolution data but also lacking in interpretability. To overcome the collective difficulties, we develop a text-to-operation model to map the vague editing language request into a series of editing operations, e.g., change contrast, brightness, and saturation. Each operation is interpretable and differentiable. Furthermore, the only supervision in the task is the target image, which is insufficient for a stable training of sequential decisions. Hence, we propose a novel operation planning algorithm to generate possible editing sequences from the target image as pseudo ground truth. Comparison experiments on the newly collected MA5k-Req dataset and GIER dataset show the advantages of our methods. Code is available at https://github.com/jshi31/T2ONet.

We investigate the weakly-supervised scene graph generation, which is a challenging task since no correspondence of label and object is provided. The previous work regards such correspondence as a latent variable which is iteratively updated via nested optimization of the scene graph generation objective. However, we further reduce the complexity by decoupling it into an efficient first-order graph matching module optimized via contrastive learning to obtain such correspondence, which is used to train a standard scene graph generation model. The extensive experiments show that such a simple pipeline can significantly surpass the previous state-of-the-art by more than 30% on the Visual Genome dataset, both in terms of graph matching accuracy and scene graph quality. We believe this work serves as a strong baseline for future research. Code is available at https://github.com/jshi31/WS-SGG.

Editing an image automatically via a linguistic request can significantly save laborious manual work and is friendly to photography novice. In this paper, we focus on the task of language-guided global image editing. Existing works suffer from imbalanced and insufficient data distribution of real-world datasets and thus fail to understand language requests well. To handle this issue, we propose to create a cycle with our image generator by creating a novel model called Editing Description Network (EDNet) which predicts an editing embedding given a pair of images. Given the cycle, we propose several free augmentation strategies to help our model understand various editing requests given the imbalanced dataset. In addition, two other novel ideas are proposed: an Image-Request Attention (IRA) module which allows our method to edit an image spatial-adaptively when the image requires different editing degree at different regions, as well as a new evaluation metric for this task which is more semantic and reasonable than conventional pixel losses (e.g. L1). Extensive experiments on two benchmark datasets demonstrate the effectiveness of our method over existing approaches.

Learning from image-text data has demonstrated recent success for many recognition tasks, yet is currently limited to visual features or individual visual concepts such as objects. In this paper, we propose one of the first methods that learn from image-sentence pairs to extract a graphical representation of localized objects and their relationships within an image, known as scene graph. To bridge the gap between images and texts, we leverage an off-the-shelf object detector to identify and localize object instances, match labels of detected regions to concepts parsed from captions, and thus create "pseudo" labels for learning scene graph. Further, we design a Transformer-based model to predict these "pseudo" labels via a masked token prediction task. Learning from only image-sentence pairs, our model achieves 30% relative gain over a latest method trained with human-annotated unlocalized scene graphs. Our model also shows strong results for weakly and fully supervised scene graph generation. In addition, we explore an open-vocabulary setting for detecting scene graphs, and present the first result for open-set scene graph generation.

Understanding web instructional videos is an essential branch of video understanding in two aspects. First, most existing video methods focus on short-term actions for a-few-second-long video clips; these methods are not directly applicable to long videos. Second, unlike unconstrained long videos, e.g., movies, instructional videos are more structured in that they have step-by-step procedures constraining the understanding task. In this work, we study problem-solving on instructional videos via Visual Question Answering (VQA). Surprisingly, it has not been an emphasis for the video community despite its rich applications. We thereby introduce YouCookQA, an annotated QA dataset for instructional videos based on YouCook2 [27]. The questions in YouCookQA are not limited to cues on a single frame but relations among multiple frames in the temporal dimension. Observing the lack of effective representations for modeling long videos, we propose a set of carefully designed models including a Recurrent Graph Convolutional Network (RGCN) that captures both temporal order and relational information. Furthermore, we study multiple modalities including descriptions and transcripts for the purpose of boosting video understanding. Extensive experiments on YouCookQA suggest that RGCN performs the best in terms of QA accuracy and better performance is gained by introducing human-annotated descriptions. YouCookQA dataset is available at https://github.com/Jossome/YoucookQA.

We invest the problem of weakly-supervised video grounding, where only video-level sentences are provided. This is a challenging task, and previous Multi-Instance Learning (MIL) based image grounding methods turn to fail in the video domain. Recent work attempts to decompose the video-level MIL into frame-level MIL by applying weighted sentence-frame ranking loss over frames, but it is not robust and does not exploit the rich temporal information in videos. In this work, we address these issues by extending frame-level MIL with a false positive frame-bag constraint and modeling the visual feature consistency in the video. In specific, we design a contextual similarity between semantic and visual features to deal with sparse objects association across frames. Furthermore, we leverage temporal coherence by strengthening the clustering effect of similar features in the visual space. We conduct an extensive evaluation on YouCookII and RoboWatch datasets, and demonstrate our method significantly outperforms prior state-of-the-art methods.

Expectation maximization (EM) algorithm is to find maximum likelihood solution for models having latent variables. A typical example is Gaussian Mixture Model (GMM) which requires Gaussian assumption, however, natural images are highly non-Gaussian so that GMM cannot be applied to perform image clustering task on pixel space. To overcome such limitation, we propose a GAN based EM learning framework that can maximize the likelihood of images and estimate the latent variables. We call this model GAN-EM, which is a framework for image clustering, semi-supervised classification and dimensionality reduction. In M-step, we design a novel loss function for discriminator of GAN to perform maximum likelihood estimation (MLE) on data with soft class label assignments. Specifically, a conditional generator captures data distribution for K classes, and a discriminator tells whether a sample is real or fake for each class. Since our model is unsupervised, the class label of real data is regarded as latent variable, which is estimated by an additional network (E-net) in E-step. The proposed GAN-EM achieves state-of-the-art clustering and semi-supervised classification results on MNIST, SVHN and CelebA, as well as comparable quality of generated images to other recently developed generative models.