Skip to content
Date and source

Dates follow the news timeline in Beijing time and include both endpoints.

Jul 23Wed
  1. Hugging Face Blog62

    Hugging Face Releases a Fast Inference Optimization for Flux LoRA

    Hugging Face releases a LoRA inference optimization for Flux.1-Dev, combining Flash Attention 3, torch.compile, and FP8 quantization to achieve a 2.23× speedup on H100.

    Why it matters: Provides a complete optimization recipe for Flux LoRA inference, including real-world data for hot-swapping and consumer-grade GPUs, which can be directly applied to your own image generation pipeline.

Jun 19Thu
  1. Hugging Face Blog62

    Hugging Face Tutorial: Fine-Tuning FLUX.1-dev with QLoRA on Consumer Hardware

    Hugging Face publishes a tutorial demonstrating how to fine-tune FLUX.1-dev on a single GPU using QLoRA and diffusers, with peak VRAM around 9GB. On an RTX 4090 at 512x768 resolution with LoRA rank 4, training for 700 steps takes approximately 41 minutes, whereas standard BF16 LoRA requires 26GB of VRAM.

    Why it matters: Hugging Face officially provides the complete workflow and measured VRAM data for fine-tuning FLUX.1-dev on single-GPU consumer hardware, which can be adapted for your own style training.

Jun 3Tue
May 21Wed
Feb 12Wed
Jan 27Mon
  1. Hugging Face Blog60

    The Diffusers Team Surveys the State of Open-Source Video Generation Models and Optimization Approaches

    The Hugging Face Diffusers team publishes an article surveying the state of open-source video generation models and introducing how Diffusers supports these models. The article lists open-source models such as CogVideoX, Mochi-1, HunyuanVideo, Allegro, and LTX-Video, noting that high resource requirements, generalization ability, and latency are the main limitations.

    Why it matters: The Diffusers team systematically surveys the state of open-source video models and provides an optimization combination that reduces HunyuanVideo's VRAM usage to approximately 6.5GB.

Oct 22Tue
Sep 30Mon
  1. Hugging Face Blog42

    How to convert vertex-colored meshes to UV-mapped textured meshes

    This tutorial introduces how to use the InstantTexture library to convert vertex-colored .obj meshes output by generative 3D models (such as InstantMesh) into UV-mapped .glb textured meshes. The core steps involve using xatlas to generate UV mappings, then filling 1024 textures using barycentric coordinate interpolation, along with patching, median filtering, Gaussian blur, and LANCZOS downsampling to eliminate holes.

Jul 30Tue
  1. Hugging Face Blog38

    Memory-efficient Diffusion Transformer with Quanto and Diffusers

    Hugging Face demonstrates how to use the Quanto quantization tool in the Diffusers library to reduce memory usage of Transformer diffusion models. Using PixArt-Sigma, Stable Diffusion 3, and Aura Flow as examples, quantizing the diffusion backbone and text encoder to FP8 reduces SD3 inference VRAM from 16.403 GB to 8.204 GB, with virtually no quality loss and a slight increase in latency.

Jan 15Mon
Jan 14Sun
Jan 2Tue
  1. Hugging Face Blog62

    Hugging Face Releases Advanced Training Script for SDXL Dreambooth LoRA

    Hugging Face releases an advanced training script for SDXL Dreambooth LoRA, integrating Pivotal Tuning and the Prodigy optimizer into diffusers, along with reusable training parameters.

    Why it matters: Hugging Face officially integrates Pivotal Tuning and the Prodigy optimizer into diffusers training scripts, providing directly reusable parameters and experimental comparisons.

Dec 5Tue
Nov 9Thu
  1. Hugging Face Blog78

    Using Latent Consistency LoRA to Generate SDXL Images in 4 Steps

    Hugging Face releases the LCM LoRA method, which trains a small number of LoRA adapter layers instead of distilling the entire model, reducing image generation for SDXL and Stable Diffusion from 25 to 50 steps down to 4 to 8 steps.

    Why it matters: LCM LoRA enables SDXL to generate images in 4 steps. The article provides directly runnable diffusers code and speed comparisons across multiple hardware configurations.

Oct 24Tue
Oct 2Mon
Sep 29Fri
Sep 18Mon
  1. Hugging Face Blog30

    Getting started with 3D Gaussian Splatting: principles and prospects of real-time radiance field rendering

    3D Gaussian Splatting is a rasterization technique that can learn from a small number of images and render photorealistic scenes in real time. It first uses COLMAP for Structure from Motion to estimate a point cloud, then converts each point into a Gaussian and trains it, achieving fast rendering through differentiable Gaussian rasterization. The costs are high VRAM usage (4GB for viewing, 12GB for training) and large scene files (1GB+), and it is not yet compatible with existing rendering pipelines such as Vulkan, DirectX, and WebGPU.

Aug 30Wed
Aug 9Wed
  1. Hugging Face Blog22

    How to optimize the Bark text-to-speech model with 🤗 Transformers

    Based on 🤗 Transformers, Optimum, and Accelerate, three optimizations can reduce the VRAM usage of the Bark text-to-speech model and speed up inference. Bark was proposed by Suno AI, has been supported since Transformers v4.31.0, consists of four models—semantic, coarse acoustics, fine acoustics, and EnCodec decoding—and offers small and large checkpoints.

  2. Hugging Face Blog22

    Deploying Hugging Face models with BentoML: a DeepFloyd IF hands-on guide

    This tutorial demonstrates how to deploy DeepFloyd IF as a production-grade service using the open-source model serving framework BentoML. DeepFloyd IF is an open-source text-to-image model that works directly in pixel space, consists of a frozen text encoder and three cascaded pixel diffusion modules, can progressively upscale a 64x64 px image to 1024x1024 px, and encodes prompts with T5-XXL-1.1.

Aug 4Fri
Jul 17Mon
  1. Hugging Face Blog60

    Hugging Face builds an AI WebTV video stream with Zeroscope and MusicGen

    Hugging Face released an experimental AI WebTV demo that uses the open-source text-to-video model Zeroscope and the music generation model MusicGen to automatically generate and stream short videos.

    Why it matters: It breaks down the complete pipeline for building an automatically generated video stream with Zeroscope, MusicGen, and FILM, and includes failure cases and prompt suggestions.

Jul 14Fri
Jun 15Thu
May 25Thu
  1. Hugging Face Blog22

    Optimizing Stable Diffusion for Intel CPUs with NNCF and 🤗 Optimum

    Using quantization-aware training (QAT) with OpenVINO NNCF combined with Token Merging, Stable Diffusion achieves a 5.1x inference speedup and a 4x reduction in model size on Intel CPUs. The optimization starts from a model fine-tuned on the Pokemons dataset, requires only 4096 iterations, completes within one day on a single GPU with 24GB of VRAM, and has been integrated into the Diffusers training scripts.

May 23Tue
Apr 26Wed
Mar 28Tue
Mar 24Fri
  1. Hugging Face Blog62

    Train your own ControlNet with diffusers

    Hugging Face released a tutorial explaining how to train ControlNet with diffusers, walking through a three-step process using the face pose model Uncanny Faces as an example.

    Why it matters: Using the face pose ControlNet as an example, it walks through the full process of conditioning planning, dataset construction, and the diffusers training script, and provides low-VRAM parameter configurations.

Mar 3Fri
  1. Hugging Face Blog62

    Diffusers integrates ControlNet: a tutorial on using StableDiffusionControlNetPipeline

    Hugging Face integrated ControlNet into Diffusers and introduced StableDiffusionControlNetPipeline, which can control Stable Diffusion generation using spatial conditions such as depth maps, segmentation maps, scribbles, and keypoints.

    Why it matters: The official Diffusers documentation provides the complete way to integrate ControlNet, including reusable code for Canny, OpenPose, and multi-condition combinations.

Jan 26Thu
  1. Hugging Face Blog73

    Hugging Face explains how to efficiently fine-tune Stable Diffusion with LoRA

    Hugging Face published a blog post on how to efficiently fine-tune Stable Diffusion with LoRA, and LoRA training support is now available in diffusers, covering both Dreambooth and full fine-tuning approaches.

    Why it matters: The official diffusers documentation provides LoRA fine-tuning and inference scripts and notes that weights can be reduced to about 3 MB, making it easy for creators to evaluate a low-cost customization workflow.

Jan 20Fri
  1. Hugging Face Blog22

    AI applications in game development: 3D asset generation

    Text-to-3D is not yet practically usable for game development; generated meshes require extensive manual processing to meet game-ready standards. Approaches such as DreamFusion, Point-E, and CLIP-Forge are mostly based on view synthesis or NeRF and are incompatible with the mesh pipelines that games rely on. In a 5-day farming game project, the author instead used differently colored cubes to represent crops.

Dec 1Thu
  1. Hugging Face Blog62

    How to run Stable Diffusion with Core ML on Apple Silicon

    Apple engineers provided conversion scripts and inference code, and Hugging Face has converted the weights of Stable Diffusion v1.4, v1.5, v2 base, and v2.1 base into Core ML versions and published them on the Hugging Face Hub.

    Why it matters: Apple engineers and Hugging Face converted Stable Diffusion weights to Core ML and provided two inference paths, in Python and Swift, for generating images locally on Apple Silicon.

Nov 7Mon
Aug 22Mon
  1. Hugging Face Blog62

    Hugging Face Explains How to Run Stable Diffusion with Diffusers

    Hugging Face releases a tutorial explaining how to run the Stable Diffusion text-to-image model with the Diffusers library and breaking down how it works.

    Why it matters: Hugging Face officially breaks down how Stable Diffusion is invoked in Diffusers and the principles of Latent Diffusion's three components, making the image generation process easier to understand.

Jun 7Tue
  1. Hugging Face Blog20

    Illustrated Diffusion Models: From DDPM to a Step-by-Step PyTorch Implementation

    This Hugging Face blog post takes an in-depth look at denoising diffusion probabilistic models (DDPM), based on the 2020 paper by Ho et al., with a step-by-step PyTorch implementation whose code references Phil Wang's implementation. The article covers both the forward noising and reverse denoising processes; the DDPM authors use T=1000 time steps, encode time steps with sinusoidal positional embeddings, and use Wide ResNet blocks as the core module of the U-Net.