Skip to main content
GameDev.net gamedev.net

Image Generation News

The latest Image Generation coverage curated for game developers.

When Images Look Right and Retrieve Wrong: Coverage-Guided Cross-Scale Re-Indexing for Knowledge-Faithful Generative Perception

A new multimodal indexing method, CERES, tackles a subtle failure mode in generative perception: images can look correct yet lose the concepts needed for retrieval. The system keeps scale-sensitive entities …

Research arXiv cs.GR · 1 month ago
FontFusion: Enhancing Generative Text in Diffusion Models with Typographic Conditioning

FontFusion tackles a real pain point for anyone generating text in images: better font control usually makes the text less readable. The plug-in framework for DiT models claims to keep …

Research arXiv cs.GR · 3 months, 2 weeks ago
Squeezing Capacity from Multimodal Large Language Models for Subject-driven Generation

A new paper argues that subject-driven image generation works better when the diffusion model is conditioned on a multimodal LLM that sees text and reference images together. The authors also …

Research arXiv cs.GR · 3 months, 4 weeks ago
Colorful-Noise: Training-Free Low-Frequency Noise Manipulation for Color-Based Conditional Image Generation

This research introduces a novel approach to image generation that allows for low-frequency noise manipulation without the need for extensive training. By focusing on low-frequency components, developers can gain better …

Research arXiv cs.GR · 4 months, 2 weeks ago
HIGS: History-Guided Sampling for Diffusion Models

The introduction of History-Guided Sampling (HiGS) marks a significant advancement in diffusion models, enhancing image generation quality and efficiency. By integrating recent model predictions into each inference step, HiGS allows …

Research Disney Research Studios · 5 months ago
Image Generation from Contextually-Contradictory Prompts

A new framework for image generation tackles the challenge of contextual contradictions in prompts. This advancement is particularly relevant for graphics programmers and artists, as it enhances the accuracy of …

Research arXiv cs.GR · 5 months, 4 weeks ago
TAUE: Training-free Noise Transplant and Cultivation Diffusion Model

The TAUE model introduces a significant leap in image generation for game developers, allowing for layer-wise control without the need for extensive training datasets. This innovation enables artists to create …

Research arXiv cs.GR · 6 months ago
Prompt-Driven Color Accessibility Evaluation in Diffusion-based Image Generation Models

The introduction of CVDLoss marks a significant advancement in evaluating color accessibility in image generation models. This is particularly relevant for artists and designers who aim to create inclusive visuals. …

Research arXiv cs.GR · 6 months, 1 week ago
Image Generation Models: A Technical History

The rapid evolution of image generation models is reshaping the landscape for graphics programmers and artists alike. This comprehensive survey highlights key advancements in techniques like GANs and VAEs, providing …

Research arXiv cs.GR · 6 months, 2 weeks ago
FontUse: A Data-Centric Approach to Style- and Use-Case-Conditioned In-Image Typography

FontUse introduces a data-centric approach to typography in image generation, addressing a common challenge developers face. By leveraging a dataset of 70K images with detailed annotations, it allows for better …

Research arXiv cs.GR · 6 months, 2 weeks ago
Navigating with Annealing Guidance Scale in Diffusion Space

A new annealing guidance scheduler enhances the performance of denoising diffusion models in text-to-image generation. This innovation allows for dynamic adjustments to the guidance scale, improving both image quality and …

Research arXiv cs.GR · 6 months, 3 weeks ago
PBR-Inspired Controllable Diffusion for Image Generation

A new pipeline for image generation allows developers to control geometric layouts and PBR material properties more effectively. By utilizing a G-buffer, users can manipulate elements within a scene, enhancing …

Research arXiv cs.GR · 7 months, 2 weeks ago
ImageRAG: Dynamic Image Retrieval for Reference-Guided Image Generation

ImageRAG introduces a significant advancement in image generation by dynamically retrieving relevant images based on text prompts. This method enhances the ability to generate rare and fine-grained concepts without requiring …

Research arXiv cs.GR · 7 months, 2 weeks ago
X2HDR: HDR Image Generation in a Perceptually Uniform Space

The new X2HDR technique allows existing image generators to produce high-dynamic-range (HDR) images without extensive retraining. This advancement is particularly beneficial for graphics programmers and artists, as it enhances image …

Research arXiv cs.GR · 7 months, 2 weeks ago
Scaling NVFP4 Inference for FLUX.2 on NVIDIA Blackwell Data Center GPUs

NVIDIA's collaboration with Black Forest Labs has led to significant advancements in the FLUX.2 model, particularly in optimizing inference performance on Blackwell GPUs. This innovation reduces memory requirements by over …

Official NVIDIA Graphics · 8 months ago
HiWave: Training-Free High-Resolution Image Generation via Wavelet-Based Diffusion Sampling

HiWave introduces a groundbreaking, training-free method for ultra-high-resolution image generation that significantly improves visual fidelity. This innovation is particularly relevant for graphics programmers and artists, as it enhances the quality …

Research Disney Research Studios · 9 months, 1 week ago
MACS: Multi-source Audio-to-image Generation with Contextual Significance and Semantic Alignment

The introduction of MACS marks a significant advancement in audio-to-image generation, allowing for multi-source audio inputs to create richer visual content. This innovation is particularly relevant for audio engineers and …

Research arXiv cs.GR · 9 months, 1 week ago
F-scheduler: illuminating the free-lunch design space for fast sampling of diffusion models

The new F-scheduler enhances diffusion models, enabling the sampling of high-resolution images in fewer steps, which is crucial for graphics programmers. This innovation allows for a 1024x1024 image to be …

Research arXiv cs.GR · 9 months, 3 weeks ago
Controllable Layer Decomposition for Reversible Multi-Layer Image Generation

The introduction of Controllable Layer Decomposition (CLD) revolutionizes multi-layer image generation by allowing fine-grained control over image layers. This method significantly enhances the editing process for developers, enabling reversible manipulation …

Research arXiv cs.GR · 9 months, 4 weeks ago