context-aware synthesis

3D scene generation suffers from limited diversity, low visual fidelity, and poor view consistency—hindering its deployment in immersive media, robotics, autonomous driving, and embodied AI. This paper presents a systematic survey of four dominant paradigms: procedural, neural 3D, image-driven, and video-driven generation. We introduce the first unified taxonomy to clarify technical evolution across these approaches. Three emerging frontiers are identified: physics-aware modeling, interactive generation, and perception-generation co-design. Leveraging NeRF, 3D Gaussian Splatting, diffusion models, GANs, and multimodal representations, we conduct a rigorous cross-paradigm evaluation on standard benchmarks, quantifying trade-offs among fidelity, diversity, and view consistency. To foster reproducibility and community advancement, we publicly release an open-source tracking platform that continuously monitors state-of-the-art progress.

Conditional image synthesis with diffusion models suffers from a lack of systematic understanding due to architectural complexity, task heterogeneity, and diverse conditional mechanisms. Method: This paper introduces the first unified taxonomy for conditional diffusion modeling, categorizing approaches by *where* conditioning is injected—either into the denoising network architecture or the sampling process—and formalizes three paradigmatic stages: training, reuse, and specialization. It further classifies six mainstream sampling-time conditioning strategies. Contributions: Based on a structured analysis of over 100 works, the paper establishes a comprehensive knowledge framework and open-sources an authoritative resource repository (GitHub Awesome-Conditional-Diffusion-Models). It identifies persistent bottlenecks—including limited generalization, inefficient inference, and coarse-grained control—and proposes principled directions toward scalable, modular, and fine-grained conditional modeling.

Existing text-to-image models lack explicit, continuous control over camera intrinsics—such as focal length and field of view—resulting in geometric and semantic inconsistencies across lens configurations and limiting applicability in professional photography. This work introduces a text-driven, photorealistic image generation framework that pioneers “dimensional lifting” and “contrastive camera learning” to enable differentiable, scene-consistent modeling of camera parameters for the first time. Technically, it integrates camera-parameter embeddings, geometry-aware attention mechanisms, and multi-scale contrastive losses into a diffusion architecture to explicitly encode physical imaging priors. Experiments demonstrate significant improvements over Stable Diffusion 3 and FLUX on lens-switching, depth-of-field, and perspective transformation tasks. Generated images exhibit both physical plausibility and high visual fidelity, advancing controllable, physics-informed generative modeling.

Existing 3D generative models neglect scene-specific constraints, resulting in synthetic assets that fail to integrate naturally into real-world environments. To address this, we propose a scene-conditioned framework for 3D object style transfer and compositing. Our method jointly optimizes object texture and environment lighting via differentiable ray tracing, while leveraging image priors from pre-trained text-to-image diffusion models (e.g., Stable Diffusion) to ensure geometric–photometric consistency and semantic adaptability in an end-to-end manner. Crucially, this work establishes the first tight coupling between 3D stylization and 2D scene semantics, enabling dynamic, context-aware relighting and restyling of a single 3D object across diverse semantic settings (e.g., summer/winter, fantasy/futuristic). We validate our approach on varied indoor/outdoor scenes and arbitrary 3D objects, demonstrating substantial improvements in visual realism, physical plausibility, and artistic controllability of composited imagery.

This paper addresses key deployment bottlenecks hindering practical adoption of generative AI for image synthesis—namely, high computational overhead, data bias, and poor alignment with user intent. To tackle these challenges, we propose a structured, input-modality–centric taxonomy that unifies modeling across GANs, diffusion models, and conditional generation paradigms. We systematically categorize core tasks—including image-to-image translation, text-to-image generation, domain adaptation, and multimodal alignment—and conduct an in-depth analysis of architectural design principles and applicability boundaries of representative models such as DALL·E, ControlNet, and DeepSeek Janus-Pro. Furthermore, we establish an industrial-deployment–oriented evaluation framework, explicitly delineating optimization pathways for computational efficiency, bias mitigation, and intent alignment. The resulting methodology provides researchers and practitioners with a theoretically grounded yet practically actionable guide for developing and deploying robust, equitable, and controllable generative image systems.

A persistent modality gap exists between 3D reconstruction and generation: reconstruction typically requires dense multi-view inputs, whereas generation often operates from single- or zero-view conditions—stemming from misalignment between geometric constraints and generative priors. This work introduces the first reconstruction-driven video diffusion framework, unifying sparse-view RGB-D reconstruction with video diffusion modeling. Key contributions include: (1) a cyclic fusion paradigm that dynamically augments training views, overcoming viewpoint saturation; and (2) RGB-D rendering-guided conditional modeling coupled with an iterative generation-reconstruction closed-loop optimization. Evaluated under extremely sparse viewpoints and severe occlusions, the method achieves significant improvements in novel-view synthesis quality—surpassing state-of-the-art methods in PSNR and SSIM—while demonstrating markedly enhanced generalization and robustness.

This work addresses the problem of generating multi-view-consistent, photorealistic color images of natural scenes from a single semantic mask, without requiring multi-view supervision or category-specific priors. Methodologically, we introduce the “Semantic Field”—a learnable intermediate representation bridging semantic masks and neural radiance fields (NeRFs)—which implicitly maps semantic layouts to 3D geometry and appearance by distilling prior knowledge from pretrained semantic-to-image synthesis models (e.g., Seg2Image). Training is fully self-supervised using only a collection of monocular Internet images. To our knowledge, this is the first framework enabling open-domain, 3D-consistent view synthesis for complex natural scenes—including intricate terrain, vegetation, and organic structures—from a single semantic mask. Experiments demonstrate significant improvements over existing baselines in both visual fidelity and cross-view consistency, producing high-fidelity, geometrically plausible videos of dynamic natural scenes.

Existing layered image synthesis methods face limitations in foreground-background separation, data availability, synthesis quality, and scene diversity. This work proposes the BFS framework, which, for the first time, transfers knowledge from non-layered image synthesis to layered generation. BFS employs a dual-branch diffusion model that jointly synthesizes a foreground layer—complete with visual effects such as shadows and reflections—and a composite image, ensuring photorealism and coherence. To address data scarcity, the method introduces a two-stage training strategy that requires only high-quality non-layered images. Experimental results and user studies demonstrate that BFS significantly outperforms current approaches in terms of synthesis quality, visual consistency, and scene diversity.

Current visual generative models exhibit significant limitations in spatial reasoning, state persistence, long-term consistency, and causal understanding, hindering their ability to produce structurally coherent and intelligently behaving content. This work proposes a paradigm shift from appearance-based synthesis toward intelligent visual generation, introducing a novel five-level generative capability taxonomy—from atomic generation to world modeling—that emphasizes the integration of structure, dynamics, domain knowledge, and causality. By leveraging key technical components including a unified understanding-generation architecture, flow matching, enhanced representations, post-training optimization, and synthetic data distillation, the study establishes a capability-centered evaluation framework. This framework exposes the prevailing overreliance on perceptual quality metrics while neglecting structural and causal deficiencies, thereby charting a roadmap for the development of next-generation intelligent visual generation systems.

Existing single-image 3D scene reconstruction methods struggle to produce editable, physically consistent textured meshes—suffering from erroneous object decomposition, inaccurate spatial relationships, and missing backgrounds—thus failing industrial requirements in film and game production. This paper introduces the first end-to-end framework for editable 3D asset generation: it enforces geometric plausibility via a novel 4-DoF differentiable ground-plane constraint; models occlusion recovery as a generative image editing task; and achieves, for the first time, background-driven spatially consistent reconstruction, yielding illumination-coherent, simulation-ready, fully textured meshes. The method synergistically integrates state-of-the-art modules—including object detection, monocular depth estimation, NeRF/3D Gaussian Splatting reconstruction, diffusion-based generation, and differentiable geometric optimization—across complementary domains. It establishes new state-of-the-art performance on single-image 3D scene reconstruction, producing structurally sound, texture-accurate, physically plausible, and post-editable 3D scenes compatible with standard pipelines.

This study systematically evaluates the capabilities and limitations of generative image restoration (GIR) methods in practical applications, revealing a shift in failure modes from under-generation to over-generation. To this end, we establish a multidimensional evaluation framework that comprehensively analyzes the performance of diffusion models, GANs, PSNR-oriented approaches, and general-purpose generative models across key dimensions including detail fidelity, sharpness, semantic correctness, and overall perceptual quality. Through large-scale subjective and objective experiments, we identify the central challenges as balancing fine-grained detail preservation with semantic controllability. Leveraging these insights, we develop a novel image quality assessment (IQA) model better aligned with human perception, offering a new benchmark and guiding direction for future GIR research.

This work addresses the challenge of structural distortion in multi-condition image generation caused by conflicting conditioning signals, which limits the utility of synthetic data for high-level autonomous driving tasks. To mitigate this issue, the authors propose an attention-based multi-condition diffusion model that explicitly suppresses condition conflicts during generation by jointly leveraging semantic segmentation maps, depth maps, and edge information. The method achieves, for the first time, high-fidelity structural preservation in multi-condition image synthesis within autonomous driving scenarios and introduces a dedicated evaluation framework. Experimental results demonstrate that the generated images significantly outperform existing approaches in structural consistency, effectively alleviating data scarcity and establishing a new benchmark for future research.