AI Briefing#diffusion-model
The latest AI and developer news about #diffusion-model, with the original source and a short summary.
The latest AI and developer news about #diffusion-model, with the original source and a short summary.
Photoroom used 32 H200 GPUs to release an optimized recipe and code for training a text-to-image model in just 24 hours.
SK Planet unpacked the structure of personalized voice generation using Zero-Shot Voice Conversion.
SP4D generates RGB and kinematic part videos together from a single-view input.
By finding material embeddings in CLIP-space, it implements material blending and fine-grained attribute control.
It generates 4D face avatars with multi-view consistency and expressions from just a single image.
SD3-Turbo, which generates high-resolution images with just 4 sampling steps, was introduced through LADD technology.
Stable Virtual Camera is a general-purpose diffusion model that generates highly consistent multi-view videos through 3D camera control.
FuriosaAI presented research papers on multimodal learning and diffusion model control at ECCV and CoLM.
OpenAI has released a system card detailing the technical architecture of its video generation model Sora and the measures taken to ensure its safety.
Implemented natural likeness using an HG-DPO-based backbone model and the Instantbooth module, validated via a photo booth at if(kakaoAI)2024.
OpenAI has unveiled sCM technology that achieves quality on par with existing diffusion models using only 2 steps of sampling.
This presents a method for efficiently reducing the memory usage of Diffusion Transformer models by leveraging the Quanto quantization tool.
This introduces the image dynamics prediction research selected as a CVPR 2024 Best Paper and LG AI Research's camera exposure control research.
To overcome the slow generation speed of diffusion models, we propose **Consistency Models**, which map noise directly to data.
By introducing a method that learns directly from data without distillation, the generation quality of the Consistency Model has been significantly improved.
OpenAI has revealed the technical principles behind Voice Engine, which generates custom voices from a 15-second audio sample, along with details of its safety research.
Improved noise sampling for Rectified Flow models and introduced a new Transformer architecture to enhance high-resolution image synthesis performance.
SegMoE, a Mixture-of-Experts (MoE) framework for diffusion models, has been released.
HuggingFace's TRL library now supports DDPO, enabling RLHF fine-tuning of Stable Diffusion models.
Würstchen, a new diffusion model that dramatically reduces training and inference costs through a high spatial compression ratio of 42x, has been released.
Marking the first anniversary of its release, Hugging Face's Diffusers library summarized its major feature updates and development over the past year.
Hugging Face has announced an ethical framework for the responsible development of the Diffusers library.
This covers the technical features and implementation of VQ-Diffusion, a conditional latent diffusion model that operates in a discrete latent space.
This article explains the mathematical principles of DDPM in detail and provides step-by-step implementation code using PyTorch.