digital content creators

Abstract: One-step and few-step text-to-image generation must accurately render the details specified in the prompt despite limited opportunities for refinement. Distribution matching distillation (DMD) uses score differences to bring the generated distribution closer to a reference distribution. Following prior work, we estimate this difference with internal velocity branches, avoiding a separate fake-score network. We ask whether an explicit moment constraint improves this learned score supervision. A decomposition of the reverse-KL field into a Gaussian moment component and a non-Gaussian residual identifies statistics that can be constrained explicitly. We introduce Intrinsic Distribution Matching (IDM), which retains the full DMD correction and matches feature means and per-coordinate standard deviations using the generator's exponential-moving-average (EMA) copy, requiring no external feature encoder. Experiments use full-parameter and low-rank adaptation (LoRA) tuning on Z-Image-Turbo 6B and Qwen-Image 20B. At two network evaluations, IDM reaches 94.25% and 90.20% on Chinese and English LongText-Bench, the highest scores among the compared methods at this budget. Controlled ablations show gains from the added moment loss, while generator EMA features perform competitively with an external DINOv2 encoder across the five reported metrics. Code is available at https://github.com/inclusionAI/IDM.

Abstract: Humans routinely consume multimodal content, where textual and visual information is naturally interleaved. Although Unified Multimodal Models (UMMs) can jointly generate text and images, multimodal interleaved generation remains largely underexplored. Progress in this area is hindered by two key limitations: (1) the lack of well-defined interleaved generation tasks and corresponding training data, and (2) the absence of fine-grained evaluation protocols that can reliably diagnose model performance across different capability dimensions. To this end, we present IntReal, comprising two complementary components: IntReal-100K, a real-world dataset of multimodal interleaved samples spanning four major categories, and the IntReal Evaluation Suite, which evaluates model performance across seven complementary dimensions using specialized expert models, enabling coarse-to-fine diagnosis. Extensive experiments validate our evaluation suite through low inter-metric redundancy (max pairwise absolute Spearman's $ρ\leq 0.42$ ), high agreement with human judgments (83.3% accuracy), and strong consistency across different expert evaluators (Spearman's $ρ> 0.9$ ), supporting the complementarity and reliability of our metrics. By benchmarking four families of UMMs, we show that IntReal Evaluation Suite reveals distinct strengths and weaknesses across models. Post-training experiments demonstrate the utility of IntReal-100K for improving interleaved generation and further show that training unimodal generation can also benefit interleaved generation. These findings suggest that interleaved generation can benefit from modality-specific optimization, rather than requiring joint text-image training at every stage.

Abstract: Image generation models have rapidly advanced, with many models now supporting generation conditioned on multiple reference images and textual instructions. Such models enable a wide range of applications, including entity fusion and attribute transfer across images. However, existing benchmarks for multi-reference image generation rely on limited evaluation dimensions and fail to capture the diverse challenges of the task. To address this limitation, we propose MDMR-Bench, a Multi-Dimensional benchmark for Multi-Reference image generation. MDMR-Bench is organized around representative capability settings and consists of multiple evaluation subsets spanning diverse dimensions of multi-reference image generation, including (1) common entity fusion and attribute transfer, (2) multi-element extraction from a single image, (3) infographic manipulation, and (4) visual reasoning-intensive generation. This design enables systematic analysis of model capabilities and failure modes across different challenge dimensions. Furthermore, to achieve more reliable and interpretable evaluation, we introduce a checklist-based evaluation protocol that decomposes generation quality into fine-grained evaluation aspects. Instead of relying solely on holistic scoring, our framework uses structured checklists composed of both global-level and atomic-level queries, enabling more diagnostic assessment of model performance. By benchmarking both open-source and proprietary models, we demonstrate that MDMR-Bench provides fine-grained insights into model performance across different dimensions. The dataset and code are publicly available.

Abstract: Distilling semantic meaning from raw image data remains a challenge in machine learning. Current self-supervised paradigms face a trade-off: joint-embedding architectures produce relevant features by aggressively discarding intra-class variance, irreversibly spatial and reconstruction. Conversely, generative models preserve spatial details but struggle to isolate semantics, as their optimization is dominated by pixel variations. To bridge this divide, we propose composite subspace flow matching (CSFM). Rather than corrupting data with unstructured isotropic noise, CSFM formulates representation learning as the variance-minimizing inversion of a deterministic degradation curriculum across spectral, color, and illumination subspaces. By bounding trajectory variance with hard temporal style truncation, a capacity-constrained encoder is driven to ignore transient style artifacts and encode persistent content. Empirical evaluations across CIFAR-10, CIFAR-100, and STL-10 show that CSFM yields competitive discriminative transferability while preserving generative synthesis across both convolutional and vision transformer architectures. By avoiding the collapse pathologies of non-generative methods, CSFM forms a smooth latent space that outperforms standard optimal transport flows on linear probing and few-shot classification, while enabling coherent synthesis and trajectory interpolation. These results confirm that steering the physical dynamics of generative flows offers a scalable paradigm for learning representations that are simultaneously linearly separable and generatively invertible.

Abstract: Complex text-to-image requests remain difficult to satisfy reliably. Verifying and repairing an image can fix an individual output, but a successful repair does not necessarily carry over to other requests. We present ImageRSI, a framework that keeps the image generator frozen and combines three components: five-dimensional verification and repair within each task, an experience library shared across tasks, and offline updating of the judge. Candidate experiences enter the library only after paired A/B validation, and the judge is updated with supervised fine-tuning and GRPO on accumulated image-judgment records and evaluated before reuse. We also introduce ImageReq-Bench, a benchmark of 961 requests built around hard cases in real-world image-generation requests, each paired with an evaluation checklist. ImageRSI achieves a Checklist Strict Accuracy (CSA) of 52.6% on MindBench (with Nano Banana Pro as the generator) and 65.66% on ImageReq-Bench, and we compare system outcomes across judge-training configurations.

Abstract: Metaphor is a crucial aspect of human cognition. As such, we would like to understand how it is represented in large language models (LLMs) and how it might be controlled. We use mechanistic interpretability techniques to identify causal subspaces in a range of LLMs covering different architectures and sizes. We show that we are able to steer output on classification tasks across different parts of speech. We are also able to steer output on generation tasks, indicating that these techniques can be used to control the metaphoricity of LLM output.

Abstract: Modern text-to-image (T2I) systems use internet-scale language pretraining for text-conditioned image generation. We ask whether language pretraining is necessary for strong T2I generation. We introduce Halcyon, a family of early-fusion multimodal autoregressive models pretrained entirely from scratch on GPIC, a dataset of 100M image-text pairs. Halcyon-0.52B outperforms LlamaGen-XL/8 (48.0 vs. 63.3 gFD) with 18.2$\times$ less pretraining compute. Further, Halcyon-1.6B outperforms JiT-T2I-XXL/8++ (35.3 vs. 41.0 gFD) with 275$\times$ less pretraining compute. Controlled experiments show that text-only pretraining does not improve T2I generation. Halcyon also learns image captioning by replacing caption dropout with I2T supervision, maintaining competitive T2I generation quality under the same pretraining budget. We show that strong T2I systems can be trained from scratch without internet-scale language pretraining. Code and models are available at https://cs.stanford.edu/~keshik/halcyon

Abstract: We study Text-and-Image-to-Image (TI2I) flow matching methods for controllable image editing. In a broader generative modeling context, Classifier-Free Guidance (CFG) is established as a key component responsible for high quality outputs. However, we argue that its role in TI2I modeling is poorly understood. Specifically, we make the surprising observation that the CFG mechanism is implicitly harmful for editing in the noisy phase of flow modeling, hindering instruction following and controllability. We propose a simple correction in the early flow steps, enabling a richer alignment-preservation tradeoff at no extra cost. The effectiveness of our approach is demonstrated on both pixel-space and latent TI2I flows.

Abstract: Image-to-text (I2T) understanding and text-to-image (T2I) generation are dual tasks in unified multimodal models (UMMs). Reconstructive post-training couples these tasks, but existing reconstruction-based methods remain perception-centric and provide limited supervision for reasoning about structure, intent, and transformation. We propose ReasonUni, a reasoning-augmented reconstructive post-training framework that extends this coupling beyond perception. Given an original image and an editing instruction, the UMM first reasons about what and how to edit and describes the target state, then generates the corresponding target image conditioned on this understanding output. To jointly optimize both modules, we introduce TreeGRPO, a tree-structured reinforcement learning method that samples multiple reasoning paths and candidate images per path, then aggregates visual rewards to update both modules under a unified objective. Experiments across multiple benchmarks demonstrate consistent gains in both understanding and generation, showing that reasoning-augmented reconstruction yields genuine mutual gains for UMMs.

Abstract: Instruction-driven image editing has reached native 4K, yet how such editors are trained remains undisclosed, and pretraining one from scratch is beyond most teams. The task itself compounds the difficulty: a single end-to-end editor cannot make precise parametric adjustments, match specialized experts, or interpret the vague instructions users actually write. We argue that 4K editing is a system problem rather than a single-model problem, and present BluePixel, which couples a native-4K editing model with an agent harness; the complete training recipe is disclosed in this paper. The model makes 4K training practical: it turns a 1K-pretrained editor into a native-4K one through progressive VAE adaptation that keeps 4K training as affordable as 1K, absorbs a long tail of vertical editing skills through continual learning, and refines the aesthetic-fidelity trade-off with reinforcement learning. Distillation via source-anchored target projection (SATP) then compresses 50 denoising steps to 4, completing a 4K edit in 8.7 s on a single Kunlunxin P800 XPU while preserving task success rate and background consistency. The harness, BluePixel Agent, makes editing controllable: it decomposes user instructions into specialist and general operations, verifies every result with the same reward model that guided training, and retries within bounded attempts. On VINS-4KEval++ (3,305 native-4K cases), BluePixel ranks second among seven models, behind only GPT Image 2 in non-edited-region consistency.

Abstract: We introduce Confidential Error-correcting Codes (CECs), a new cryptographic primitive that enables reliable decoding from perturbed codewords while preserving message confidentiality. We formalize this primitive through three security properties: confidentiality, robustness, and soundness. To realize this goal, we develop a general framework for constructing CECs based on a Randomness-Security Separation (RSS) structure, which first encodes a nonce to introduce randomness and then masks the encoded payload with a pseudorandom function. Based on our RSS framework, we instantiate CECs, called Conc-CEC, using concatenated Reed-Muller and Reed-Solomon codes. The resulting Conc-CEC admits provable guarantees for all three properties. Notably, compared to PRCs (CRYPTO'24), another recent cryptographic primitive with error-correcting capability, CECs achieve significantly stronger concrete security guarantees under practical parameter settings. As a primary application of CECs, we further present CECMark, a watermarking scheme for AI-generated images. Our experiments on real-world generative image models show that CECMark consistently outperforms state-of-the-art watermarking methods, including Gaussian Shading (CVPR'24), PRCMark (ICLR'25), and T2SMark (NeurIPS'25). In particular, CECMark achieves higher robustness across all 56 evaluated perturbations and ensures a false positive rate below $10^{-5}$ without introducing side effects. Taken together, our work lays a novel cryptographic foundation for trustworthy watermarking schemes that, for the first time, unify provable security guarantees with strong practical reliability.

Abstract: Generative AI gives audiences new ways to portray real sporting figures through familiar fictional worlds. Expanded opportunities to create, however, need not broaden cultural representation or distribute attention evenly. We examine this tension through the supply, demand, and circulation of anime style football videos on Instagram. Using 2,021 posts and associated video and account data, we combine caption entity extraction, human-reviewed visual identity clustering, engagement analysis, and content-reuse networks. Representation is uneven across athletes, while Blue Lock, Naruto, and One Piece account for 60.1% of explicit franchise mentions. Among posts referencing generative tools, rankings by total engagement differ from rankings by average engagement per video, distinguishing aggregate visibility from response to individual posts. Reuse is extensive: 772 of 1,548 evaluated videos belong to 124 detected groups, including 71 spanning multiple known accounts. Shared material connects a large account component alongside smaller groups, and later observed uploads receive lower median normalized engagement than first observed appearances. These findings characterize generative cultural crossover as a practice organized around concentrated cultural resources, uneven attention, and extensive recirculation. More broadly, they distinguish opportunities for visual authorship from the processes through which particular representations become widely visible.

Abstract: Co-speech gesture video generation aims to synthesize realistic, audio-driven portrait videos from a single reference image and a speech signal. Despite advances in visual quality, diffusion-based approaches still suffer from degraded local details and imperfect audio-visual alignment. We present PAGE (Parts-Aware GuidancE), a general, backbone-agnostic optimization framework for diffusion-based generation that enhances local spatial fidelity and global audio-visual consistency at both the training-objective and denoising-trajectory levels. During training, we propose Parts-Specific Reweighting (PSR), which adaptively increases supervision in semantically important regions such as lips, eyes, and hands, improving visual fidelity in perceptually critical areas. During inference, we introduce Consistency Classifier Guidance (CCG), an audio-conditioned latent classifier that identifies and corrects audio-visual inconsistencies in the generated video. We further design Differential Guidance (DG), an efficient inference strategy that refines facial and body motion separately while avoiding cross-region interference. To enable large-scale evaluation, we construct the Chinese Announcer Speech (CAS) dataset, the first comprehensive Chinese benchmark for co-speech gesture portrait video generation. Compared with the widely used PATS dataset, CAS includes substantially more identities, video clips, and scene diversity. Extensive experiments on both PATS and CAS demonstrate that PAGE achieves state-of-the-art performance in perceptual quality, gesture diversity, and audio-visual alignment. Project page: https://aigc-explorer.github.io/PAGE/.

Abstract: Video generation supports digital content creation, with applications in virtual avatars, AI-driven filmmaking, interactive game engines, and world models. State-of-the-art systems achieve controllability primarily through image-to-video (I2V) generation, in which an injected anchor image determines the identity, composition, and style of the resulting video. In modern pipelines, this anchor is synthesized by a text-to-image (T2I) generator. This raises a largely overlooked question: Are images obtained directly from the T2I model truly suitable for current video models? Our analysis shows that anchor aesthetic and quality scores are weakly correlated with downstream video quality. The structure-and-texture score is negatively correlated with I2V video quality in both the visual and motion dimensions. Fine-tuning the T2I model on an image reward accordingly yields anchors with visibly denser structural and character detail: they score higher on image metrics, yet lower the video reward on every I2V model we test and change downstream motion unpredictably. We therefore introduce Golden Anchor Learning (GAL), to our knowledge the first framework for learning golden anchors by optimizing a T2I generator with downstream video feedback. GAL evaluates candidate anchors through the videos they induce under a frozen I2V model and converts video rewards into image-level preferences for DPO fine-tuning of the T2I generator. On all four I2V models, GAL raises the VBench average by $2.97\%$--$8.46\%$ over the untuned generator. The rendered clips also score higher under VideoReward-Overall, and their extracted frames under HPSv3 and UniPercept. GAL yields consistent motion gains across all four I2V models, as measured by a $5.62\%$--$38.24\%$ increase in VBench dynamic degree. Although GAL is trained only on HunyuanVideo-1.5 renders, these gains extend to the other three I2V models.

Abstract: Prompt learning is a popular method for adapting foundation models, but learned prompts are typically task-specific and fail to generalize to new classes, domains, or compositions of tasks. In this paper, we introduce a Diffusion Meta-Prompt (DMP) model , a framework that models the distribution of learned prompts using diffusion models. Given a repository of previously learned prompts, DMP is trained and sampled without access to the original task examples or task losses, and synthesizes new prompts conditioned on natural language task descriptions. To improve the sampling stability, we introduce a test-time steering strategy for DMP, which uses the best training-selected prompt in the repository as a latent anchor during diffusion sampling, without retraining the DMP or accessing test classes. DMP improves generalization across classification, retrieval and text-to-image generation tasks, supports concept composition and negative prompting without explicit training. It reduces storage and inference costs by over 90% compared to prompt retrieval methods. For composite classification, DMP achieves upto 2.0% average gain over prior meta-learning methods across 55 pairs of datasets with gains as high as 8.5% on specific pairs such as Eurosat and Flowers. DMP also enhances cross-task generalization with ~2-9% improvement for hierarchical classification task. We further provide a theoretical guarantee bounding the expected task loss of prompts sampled from a DMP. Code is available: https://github.com/DeepakSridhar/dmp

Abstract: Multimodal Diffusion Transformers (MM-DiTs) jointly process visual and textual representations throughout generation. These models repeatedly update the text tokens through multimodal attention, forming dynamic contextual tokens whose function is not well understood. In this work, we introduce a framework for reading this contextual space through natural-language interrogation. We train a lightweight bottleneck network that maps intermediate contextual tokens into the input space of a frozen Large Language Model (LLM), allowing the LLM to answer questions about the emerging image directly from these hidden representations. Our reader reveals that contextual tokens encode a rich, global representation of the emerging scene: generation-specific semantics, including attributes left underspecified by the prompt, are accessible surprisingly early in denoising, while increasingly fine-grained details become readable over time. Remarkably, this information remains decodable even when the MM-DiT receives an empty prompt, showing that contextual tokens accumulate substantial image-specific information from the evolving visual representation itself. We further find that generations with more readable contextual representations tend to receive higher human-preference scores. Building on these observations, we introduce Contextual Alignment, a training technique that explicitly reinforces the visual-semantic information encoded in the contextual tokens, improving generation quality and distributional coverage. Together, our results establish contextual tokens as both an interpretable view into the internal dynamics of MM-DiTs and an effective target for improving generative models.

Abstract: Diffusion models generate samples by learning to reverse a fixed corruption process, and classifier-free guidance (CFG) is the standard mechanism for conditioning this process on a desired class or prompt. CFG can be applied at varying guidance strengths, and while higher strengths improve image quality and conditional alignment, too high a guidance strength can degrade image quality and diversity. Furthermore, CFG violates principled diffusion sampling dynamics, and existing explanations for why it works despite the violation disagree on the underlying theory or do not extend to deterministic samplers used in practice. We address both these issues. We first frame CFG sampling as a continuous-time optimal control problem, treating the sampling trajectory as a sequence of controls chosen to maximise the probability of the desired condition. Solving the resulting Hamilton--Jacobi--Bellman equation shows that CFG is recovered under specific path costs when using an unconstrained control set. We argue this lack of constraint is responsible for CFG's failure at high guidance strengths, since it permits the sampling path to move arbitrarily far from the current image estimate. To fix this, we propose curvature-aware CFG (CACFG), which constrains the control set to a hypersphere informed by the Gaussian regularisation used when training variational autoencoders. We show that the control inputs produced by CFG sampling routinely violate this bound, and that across diffusion models, datasets, and guidance schedules, CACFG achieves superior generative quality at mid-to-high guidance strengths with a less severe quality-diversity tradeoff than regular CFG.

Abstract: AI-generated videos are rapidly increasing in volume, duration, and resolution, creating growing demands for efficient storage and transmission. Unlike natural videos captured from the physical world, AI-generated videos are samples from a learned generative distribution, where semantic structures are critical to content consistency, while many local textures and stochastic details can be plausibly regenerated. This distinction suggests that compression should preserve semantically important spatiotemporal information rather than reconstruct every pixel of a particular generative sample. Beyond reconstruction, AI-generated videos also create a practical need for prompt-based editing, where users expect to modify generated content while preserving its original spatiotemporal semantics. Motivated by these observations, we propose a unified compression and editing framework for AI-generated videos that incorporates a frozen video generator as a reusable generative prior. Within this framework, we design three spatiotemporal semantic importance-guided techniques that respectively address what to transmit, how much to transmit, and how to use the transmitted side information. First, an innovation selection method projects the latent discrepancy using spatiotemporal semantic importance, so that the selected innovations prioritize semantic invariants over replaceable generative variations. Second, a frame-adaptive bit allocation method estimates the nonuniform semantic demands of latent frames and allocates more innovations to frames requiring stronger semantic preservation. Third, a unified reconstruction and editing method continuously adjusts the influence of the transmitted side information, enabling the same compressed representation to provide strong guidance for faithful reconstruction or serve as a flexible semantic anchor for structure-preserving prompt-driven editing.

Abstract: Recent text-to-image generation models have achieved remarkable visual quality, but improving them through post-training remains challenging because no single reward signal captures the full range of human preference. In this work, we develop a simple and effective post-training recipe for open-domain text-to-image generation based on the composition of complementary reward signals. Our reward system consists of two main components: a preference reward, trained on large-scale human preference data using a Bradley-Terry objective to capture overall human aesthetic and perceptual preferences, and rubric-based rewards, which explicitly evaluate prompt faithfulness and other desirable properties while providing safeguards against reward hacking. A key challenge is how to combine these heterogeneous reward signals. We show that a naive weighted average leads to suboptimal optimization behavior, and propose a simple reward composition strategy that more effectively balances preference optimization with rubric satisfaction. In the Arena text-to-image leaderboard (https://arena.ai/), our RL-trained Flux2dev achieves an Elo rating 69 points above the base model, and our post-trained Ideogram-4 surpasses every open-source model on the leaderboard, reaching an Elo of 1223.5. (Claims of state-of-the-art performance are based on the Arena leaderboard snapshot as of September 4, 2026.) Our results suggest that effective rewards for frontier generative-model training require broad coverage of user intent and robustness to exploitation under optimization. To support reproducible research, we release Arena-T2I-Training, a 1K subset of training data that recovers some gains of full-scale training, providing a resource that we hope will facilitate future work on post-training for text-to-image models.

Abstract: Diffusion transformers (DiTs) generate high-fidelity images from text and image conditions, but their outputs carry large variance and their faithfulness to a desired target depends heavily on how the condition is supplied. We present DAGS, a lightweight, attention-free, disentangled appearance and geometry conditioning scheme that steers a frozen image DiT to produce high-fidelity, highly faithful, and independently controllable renders. Two small convolutional encoders compute conditioning features once per frame and inject them as a learned, per-layer, element-wise residual into the image tokens, avoiding the quadratic cost of stacking conditions through attention. Because control and temporal handling live outside the frozen backbone, we retain its vast pretrained prior and eliminate backbone-overfitting risk. We further add a small recurrent lighting stabilizer and a training-free temporal guidance term that, coupled with our conditioning, elevate a per-frame image model into a streaming renderer. DAGS produces controllable, high-quality renders at a fraction of the compute of path tracing; it is not real-time, trading compute for controllability and quality. On a matched 1-spp + G-buffer input, per-frame DAGS reconstructs +8.6 dB / +10.1 dB PSNR over the real-time denoiser Intel OIDN and the diffusion renderer RGB<->X while being 2.5-8x more temporally stable perceptually (temporal-LPIPS flicker).

Abstract: Although diffusion models produce high-quality images, they also reproduce and amplify demographic imbalances in their training data. Debiasing their generation process post-training w.r.t. some sensitive attribute usually relies on classifier guidance or explicit text extra-conditioning, but this reduces methods' applicability and output diversity. Conversely, methods promoting diversity alone do not ensure fair attribute representation. In this paper, we propose a method tackling fairness and diversity jointly that is generally applicable to any diffusion model and any sensitive attribute. To this end, an adapter connects the frozen diffusion model to a pretrained vision-language embedding space, enabling fairness and diversity guidance without sensitive-attribute annotations. For fairness, pairs of text prompts define attribute directions which guide batch composition towards specific proportions. For diversity, we introduce a score measuring disagreement between the semantic estimates derived from this representation. The formulation supports unconditional and text-conditional diffusion models, while requiring no prior knowledge or data of sensitive attribute. Experiments confirm that our method improves quality and diversity scores at comparable fairness levels.

Abstract: Standard 3D Gaussian Splatting (3DGS) learns geometry and appearance jointly from RGB supervision, making it difficult to isolate how luminance and chroma contribute to the learned representation. We study this by training models under different channel supervision, freezing their non-appearance parameters (position, scale, rotation, and opacity), and re-estimating appearance with the same solver before comparing held-out reconstruction. Across eleven benchmark scenes with four independent runs each, geometry learned from luminance alone supports held-out reconstruction 0.085 dB below RGB-trained geometry on average. If chroma is deleted from a trained model, a sufficiently expressive solver can re-fit it on the frozen geometry to the original quality or slightly better. Higher-order spherical harmonics contribute much more reconstruction quality to luminance than to chroma, improving PSNR by 1.44 dB versus 0.19 dB on average, although on mirror-like surfaces hue does still change with viewpoint. The luminance advantage is even larger when geometry is being formed. Chroma-only supervision produces geometry 3.9-5.5 dB worse than luminance-only supervision after the same appearance solve; densification explains part of this gap. Overall, geometry formation in standard 3DGS is strongly luminance-dominated but not luminance-exclusive, and much of the chromatic appearance can be recovered after spatial support has formed.

Abstract: In this paper, we propose Gestalt, a new paradigm of large multimodal model built around multimodal interplay. Despite rapid advances, large multimodal models are reaching a bottleneck: existing approaches focus primarily on accommodating additional modalities while overlooking the distinct characteristics of each modality and the relations among them. Motivated by the multistage property of human multisensory perception, we propose a multimodal interplay pyramid that organizes multimodal modeling as a progression from modality-specific processing, through cross-modal alignment, to deeper multimodal integration. Guided by this pyramid, Gestalt adopts a unified discrete diffusion framework and an interplay-partitioned architecture, with learnable interplay tokens mediating cross-modal exchange and integration. The pyramid also structures its data organization and training strategy. Strong performance across image generation, multimodal understanding, and text-only evaluation shows that Gestalt significantly improves cross-modal integration while preserving modality-specific information, effectively harnessing the strengths of diffusion-based multimodal models and offering a promising path toward unified multimodal intelligence.

Abstract: As state-of-the-art text-to-image flow models achieve near-photorealistic quality, controlling their outputs, e.g., suppressing harmful content while promoting benign alternatives, has become a central challenge. The current steering paradigm consists of adding a global steering vector to selected activations. While functional, a fixed and example-agnostic vector applied uniformly along the entire trajectory cannot adapt to the changing state of the generation and often causes unintended global changes. We introduce Steering Fields, a generalization of steering vectors that adaptively re-estimates the steering direction at each step of the generative process. Steering Fields operate on the noisy states of flow models, expose a continuous trade-off between steering strength and content preservation, and are compositional, enabling the simultaneous induction and inhibition of concepts, setting a new state of the art on safety steering benchmarks. Despite using no explicit spatial masks or object priors, the trajectory-adaptive estimation naturally preserves local structure, in a manner reminiscent of image editing. In fact, Steering Fields can serve as a structure-preserving image-editing technique that achieves state-of-the-art semantic fidelity (CLIP, VQAScore), while remaining model-agnostic and inversion-free.

Abstract: Text-to-image (T2I) generation is gaining increasing popularity with the general public, motivating the development of reliable mechanisms for copyrighting such models given their expensive training costs. An adversary may obtain and reuse a pretrained T2I model without authorization, and then serve a modified version through an API service. Such modifications may arise from ordinary downstream adaptation or deliberate attempts to erase ownership, including input-prompt preprocessing, model fine-tuning, and output post-processing. From the model owner's perspective, a key challenge is therefore to embed trigger data that remain persistent under such changes while preserving the model's normal image-generation capabilities. In this work, we propose a contrastive-style watermarking objective with a term that explicitly encourages the watermarked model to behave differently from the original model on trigger inputs. Experiments show substantially stronger trigger-data persistence than prior methods across a wide range of downstream modifications and deliberate attempts to weaken the watermark, resulting in higher detection rates, often approaching 100% TPR@FPR<$10^{-4}$.

Abstract: One-dimensional (1D) variable-length visual tokenizers enable adaptive compression by varying the number of tokens, allowing downstream autoregressive (AR) models to flexibly trade off generation quality against computational cost using a single tokenizer. However, existing approaches based on nested dropout often fail to fully exploit the representational capacity of the tokenizer, resulting in suboptimal performance in both image reconstruction and generation. In this work, we introduce NesTok, a nested self-alignment framework tailored to dynamic visual tokenizers. NesTok introduces cross-length training, which jointly optimizes reconstruction across token lengths while using the full-length sequence to guide shorter counterparts, enabling shorter token sequences to approach the reconstruction quality of full-length sequences. On ImageNet, NesTok improves substantially over standard training and achieves an rFID score of 0.98. On downstream image generation, it achieves the state-of-the-art gFID score of 1.46 on ImageNet 256$\times$256 among existing variable-length autoregressive image generation methods. Code will be available at https://github.com/jaiwei804/NesTok.

Abstract: Text-to-image users often provide concise and underspecified prompts, whereas generative models benefit from detailed textual conditions for reliable instruction following. Existing systems bridge this gap with Prompt Enhancers (PEs) that rewrite raw prompts at inference time, introducing additional latency and leaving prompt elaboration external to the generator. We instead view enhanced prompts as privileged training information and ask whether their benefits can be internalized. We propose Prompt-Enhanced On-Policy Self-Distillation (PE-OPSD) for text-to-image flow-matching models. During training, a raw-prompt student follows its own generation trajectory, while an enhanced-prompt teacher provides vector-field targets at the states visited by the student. This on-policy supervision distills the behavior induced by enhanced prompts into the raw-prompt student without requiring additional text--image pairs. At inference, both the PE and teacher are removed, and the student generates directly from raw prompts. Across multiple model families, PEs, and benchmarks, PE-OPSD achieves the strongest aggregate prompt fidelity among the evaluated baselines, yields positive aggregate visual appeal gains, and retains the base-model inference efficiency.

Abstract: Precise instruction following in image generation, such as satisfying object counts and spatial relations, remains an open challenge at least in part because it is learned using unreliable reward models such as object detectors and vision-language models. We introduce Verifiable Visual Rewards (VVR), the first framework for programmatically verifiable image rewards, and show that training on it generalizes to natural prompts. Each VVR task is a scene of geometric objects and relations among them, from which we derive both the prompt and a deterministic verifier, so tasks can be generated in any number and at any chosen complexity. We release VVRBench, with 10,000 tasks over 32 constraint types, and VVRBench-Challenge, with 720 more complex tasks; the strongest model we evaluate---GPT-Image-2.5---solves 21.4% of VVRBench-Challenge. Using VVR scores as rewards for reinforcement learning (RLVVR) raises the accuracy of Stable Diffusion 3.5 Medium on VVRBench from 2.8% to 28.3% and demonstrates consistent easy-to-hard generalization. These gains extend to out-of-domain benchmarks, and mixing VVR into existing objectives further improves overall performance and human preference, motivating the adoption of VVR into standard image generation post-training recipes.

Abstract: 2D Gaussian-based image representation is becoming increasingly popular, and our work proposes a new approach to steganography by embedding information within Gaussian parameters rather than image pixels. We first fit the parameters of this representation to the target image and employ a secret key to select a subset of these parameters for fine-tuning, allowing us to embed an 8-bit message while maintaining high visual fidelity in the reconstructed image. Thirty fitting experiments on three synthetic images show that the correct key can recover the message without error, while decoding with incorrect keys yields a Bit Error Rate (BER) of $0.543$, close to random guessing. Compared with random selection using the key, selecting the least-disturbing edits recovers the message more reliably (one-sided $p=0.031$), and the average PSNR cost is only $0.091$ dB in visual quality. The embedding method transfers to 112 natural images at $256\times256$ using 4,096 Gaussians. The correct-key BER is $0.000$, and decoding under a wrong key stays close to random guessing at $0.520$. Our method embeds the payload through three Gaussian parameter types: log-anisotropy, opacity, and color luminance. In a separate nine-fit reduced setting, color luminance is removed, so the payload uses two instead of three parameter types, a $33.3\%$ reduction; all 256 Gaussians remain in the fitted representation. The correct key still recovers the message without error. However, wrong-key BER rises from $0.514$ to $0.571$, moving farther from random guessing ($0.5$).

Abstract: Generating Scalable Vector Graphics (SVG) code from natural-language instructions is an open-ended task without absolute visual ground truth, leaving both evaluation and policy optimization without a faithful signal. Scalar metrics (CLIP, Aesthetic) calibrated on natural images transfer poorly to stylized vector content, and reusing them as RL rewards triggers reward hacking. We address both limitations with rubric-based scoring. We first establish empirically that prompting a vision-language judge with a multi-axis rubric correlates with human judgments far better than scalar metrics, both across samples and within instructions. Building on this finding, we introduce RULER (Instance-aware Rubric Rewards for Reinforcement Learning), which converts each instruction into an instance-aware rubric of six items spanning semantic, visual, and stylistic axes; a judge VLM scores rendered rollouts item-by-item, and the weighted satisfactions form a fine-grained reward optimized via Group Relative Policy Optimization. Because the rubric is derived from text alone, RULER requires neither paired SVG ground truth nor human preference labels. On MMSVG-Illustration and MMSVG-Icon, RULER lifts the rubric score from 0.432/0.395 to 0.693/0.683, surpassing dedicated SVG specialists and matching the substantially larger DeepSeek-V3, with ablations identifying rubric design as the active lever for RL on open-ended SVG generation. The project page is available at https://hangyuran.github.io/RULER/.

Abstract: 3D assembly is fundamental to modern manufacturing and digital content creation. In this paper, we present PART, a unified transformer-based framework for 3D part retrieval and assembly: given a target shape and a part library, PART automatically selects the appropriate parts and predicts their 6-DoF poses to reconstruct the target. While prior work has achieved impressive progress on assembling a pre-defined set of parts, this more practical retrieval-based setting remains largely unexplored. The task faces three key challenges: (i) a combinatorially explosive search space that grows exponentially with library size; (ii) variable-length outputs, as different targets require different numbers of parts; and (iii) continuous 6-DoF pose estimation for part assembly. To address these, we formulate retrieval and assembly as a set prediction problem and design a novel transformer-based framework that retrieves parts and regresses their poses with variable-length output. Additionally, we exploit the duality between part pose estimation and target segmentation through joint training and a novel segmentation-enhanced optimization module. Finally, We curate a large-scale dataset of 80K+ shapes, and the results show that PART generalizes to scene layouts, image targets, and real-world scans. Project Page: https://iambrc.github.io/PART-project-page/.

Abstract: Although text-to-image diffusion models generally exhibit strong prompt-following ability, we identify a persistent and previously underexplored failure pattern in which a small subset of prompts differing only in the object consistently fails to realize the same target concept under identical generation settings. We term this phenomenon object-dependent concept brittleness. Such cases suggest systematic internal blind spots rather than random sampling noise. In this paper, we present an interpretability-oriented framework to audit and minimally correct these failures. Our key idea is to analyze denoising trajectories in a step-wise sparse autoencoder (SAE) space, where abstract style and attribute concepts become more separable than in the raw denoising representation. This sparse space enables us to compare successful and failed generations, identify concept dimensions whose evidence is missing, weakened, or temporally delayed, and construct class-level concept prototypes from reliable class-consistent samples. Based on this audit process, we introduce a lightweight inference-time correction strategy that interpolates denoising features toward the corresponding prototype in SAE space. Rather than serving as a task-specific retraining method, this intervention acts as a validation of the diagnosed concept deficiency. We evaluate the proposed framework on style and attribute failure cases across multiple diffusion backbones, with significant improvements in concept consistency, text fidelity, and repair success. Further analyses show that deeper denoising representations provide clearer concept structure, while early-stage intervention offers the strongest correction leverage. Code is available at https://github.com/Metecade/Object-Dependent-Concept-Brittleness.