AI Briefing#text-to-image
The latest AI and developer news about #text-to-image, with the original source and a short summary.
The latest AI and developer news about #text-to-image, with the original source and a short summary.
The model was trained on 610M samples at 512px and 50M samples at 1024px before post-training with an FDFO variant.
Linum AI releases P-JiT, a 2B pixel-space text-to-image model trained for $21,038, which removes the VAE to achieve 4x attention context reduction.
The model features a 50M parameter denoiser and generates images in 11 seconds at 20 steps on an Intel i5 7200U.
The model achieves top rankings in internal evaluations against larger open and closed competitors while running on consumer GPUs with as little as 8GB of VRAM.
The model generates recognizable 64x64 images without using oversized text encoders or VAEs.
Linum AI proposes Pyramid-JiT, a VAE-free pixel-space text-to-image model that achieves an FD-DINOv2 score of 27.9, approaching the real-image floor of 25.4, while requiring 11.3x fewer training samples than Linum v2.
Applying FLUX.2 VAE and Muon improves training efficiency and strengthens understanding of Korean instructions
The 8B parameter model achieves competitive results with Qwen-Image 20B on specific benchmarks while using a single backbone for both generation and understanding.
The JiT-DDT architecture, adopting a pixel-space encoder-decoder structure, achieved faster convergence and improved image quality while reducing GPU usage by 3.6x compared to Linum v2.
Jasper Research released an open-source cookbook covering the principles to implementation of text-to-image models, along with a 100 million image dataset.
An open-source benchmark evaluating 52 text-to-image models with 192 prompts has been released.
MAI-Image-2.6 secured 2nd place on the Arena text-to-image leaderboard, surpassing models from Google, Meta, and xAI.
xAI has launched Imagine Image 2.0, offering high-quality image generation and editing capabilities in Grok.
SpaceXAI has released Image 2.0, enhancing precision generation and editing capabilities.
SenseNova has released a preview of U1.5 Lite, which strengthens 4K generation and precise image editing.
From the emergence of DALL-E through VQ-VAE-based models to LG AI Research's L-Verse, we look at the evolution of Vision-Language models.
LG AI Research has unveiled the architecture of the EXAONE multimodal model, which understands text and images bidirectionally.
Following NLP's success, the vision-language field is also accelerating large-scale model training leveraging massive data and infrastructure.
This explores the latest trends in personalized image/language models and NeRF technology through NVIDIA GTC 2022.
This post analyzes the research trends in architecture and downstream tasks for Score-based Generative Modeling presented at NeurIPS 2023.
This piece looks at research trends in the architecture and downstream tasks of score-based generative models presented at NeurIPS 2023.
Qwen Image Agent has dramatically improved text-to-image generation performance through planning, reasoning, and search capabilities.
The Sana 1.6B text-to-image model was quantized to 1.58-bit ternary, reducing its size by 8.6x.
The Boogu-Image-0.1 series of open-source image generation models, specialized in text rendering and image editing, has been released.