Skip to content
Date and source

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

Sep 18Fri
  1. VibeClips · AI视频创作教程22

    VIBE Tutorial: How to Create AI Digital Human Talking-Head Videos with Talking Avatar

    VIBE's Talking Avatar model requires only a frontal photo and a voice recording to generate a 480p or 720p lip-synced digital human talking-head video, with no text prompt needed, and the output language follows the recording. The model is billed per use: 35 tokens for 480p and 60 tokens for 720p, regardless of recording length, and the output aspect ratio is locked to that of the uploaded photo.

  2. WellSaid · 官方博客22

    How to Localize Marketing Campaigns with AI Voice

    WellSaid's official blog published a guide on how to localize marketing campaigns with AI voice. The process includes preparing localized scripts, selecting voices with regional accents and dialects for each market, generating and previewing voiceovers, adjusting pronunciation and pacing, and then reusing the same creative direction across markets. AI voice lets teams avoid rescheduling recording sessions for every script change, shifting time and budget from coordinating recordings to comparing and polishing versions.

  3. 优设 · 创作者专栏34

    How to make AI posters look high-end? 4 pitfalls to avoid to escape a cheap tech feel

    AI posters look cheap mostly because of four basic problems: style words that are too vague, a messy information hierarchy, too many decorative elements, and unclear prompts. The article breaks down these common issues and offers fixes: control the amount of information, establish clear primary and secondary elements, make "high-end feel" concrete, and write poster prompts in order.

Sep 17Thu
  1. Creator Animation · AI影片制作教程12

    How independent creators can sell AI films online in the first 90 days: a practical guide

    Creator Animation released a practical guide for independent creators selling AI films online in the first 90 days, splitting the release cycle into three phases: packaging and distribution infrastructure, promotion that connects directly with audiences, and retention and scaling. The platform offers the first 10GB of storage free, $0.40/GB beyond that, creators keep an 80% revenue share, viewers can watch without registering, and they can use a $9/month Download Pass or buy offline versions individually.

  2. note · howrass_note · 图像视频创作分享40

    A practice log of making a 5-minute, 2K music video with lyrics using MiniMax H3

    A video-making novice used the locally runnable MiniMax H3 video generation model, together with Google Colab cloud GPUs, to complete a music video for the SUNO original song "ヘッドホンの息": about 5 minutes 16 seconds long, 1440p (QHD), with lyrics throughout. The production ultimately used 44 storyboard images, generated more than 200 images in total, and had about 250 failed videos; the author says technical details will be covered in a separate article.

  3. ArchiGen AI · 图像生成与ComfyUI实战28

    A reusable ComfyUI workflow should withstand four different types of architectural input

    A ComfyUI workflow that can generate a single architectural rendering may just be tuned to one screenshot; ArchiGen suggests testing portability with four clearly different inputs: a clean clay look, a glass-curtain-wall interior, a low-contrast dusk view, and a cluttered urban environment, which respectively expose problems such as fabricated facade details, depth errors, fake openings, and noise leaking into structural control.

  4. VibeClips · AI视频创作教程22

    VIBE free AI video generator hands-on: what the 2026 free version can do

    VIBE's free version offers four models: Seedance Pro Fast, LTX 2 Distilled, PRUNA V and LTX 2.5 Fast, supporting text-to-video and image-to-video, with no credit card required and no trial period. Free models max out at 720p (Seedance Pro Fast is 480p), every export carries a watermark, and only PRUNA V supports audio and photo lip-sync digital human mode.

  5. ArchiGen AI · 图像生成与ComfyUI实战22

    Don't judge an AI renderer by the first image alone: ArchGen proposes an exit test for revisions, archiving and handover

    ArchGen AI proposes that procurement testing for AI renderers should not stop at image quality, speed, price and integration, but extend to revisions, handover and archiving after the first image. Its five-part exit package requires traceable sources, generation recipes, result specifications, review records and limited revision capability, and has a second operator make one small change using only the package, recording the time taken and the scope of the change's impact. The archiving test has a third person reopen the package after the project has cooled down to confirm that delivered images map to source versions, products, engines, dates and settings.

  6. MattVidPro · YouTube 创作视频50

    YuE2 local AI music model hands-on: comparison with Suno V6 and Pinokio installation tutorial

    YuE2 is the best-performing locally run AI music model to date; the video compares it with Suno V6 on songs and demonstrates exaggerated prompts and parameter adjustments. YuE2's code is under the Apache 2.0 license and its model weights are CC BY-NC 4.0; individual creators can commercialize outputs, while enterprise commercial use requires a separate license. At 17:41 there is a tutorial for installing it locally on a PC via Pinokio and Wan2GP.

Sep 16Wed
  1. note · lpp · 图像视频创作分享38

    Running YuE2 AI music generation on an RX9070XT: ROCm installation tutorial plus a bonus GUI

    Installing the YuE2-3B AI music generation model via ROCm on a Windows PC with an RX9070XT; the author published the installation steps and bonus Gradio GUI code. Switching to ROCm10.x greatly improved generation speed, while ROCm7.x requires applying a pull request on GitHub to fix a Flash Attention error. A cover-integrated GUI version was also released in the fourth week of September.

  2. Promptsref.com · 创作分享22

    3 illustration SREF codes to make Midjourney works feel more hand-drawn

    Promptsref.com shares 3 Midjourney illustration SREF codes, respectively featuring sparse, imperfect line art; rough, doodled geometric shapes; and a gothic dark style blending Japanese manga with retro posters. All three build a hand-drawn feel with black lines, warm paper tones, and a small amount of accent color, suitable for children's books, indie comics, posters, stickers, and game key art. Two of them are member-exclusive SREFs.

Sep 15Tue
  1. 张小闲 · 创作者专栏62

    Generate poster title art text with a single prompt, covering styles such as brush calligraphy, jelly letters, and neon.

    The author put together a playbook for generating poster title art text with a single prompt, writing the theme, font feel, color, and atmosphere into the prompt to get a first version out in just over ten seconds. The author tried directions including handwritten brush calligraphy, pink-orange jelly letters, purple flowing light, tech speed letters, futuristic neon, and premium gold lettering, covering scenarios such as e-commerce posters, event key visuals, product detail pages, and AI design covers; the prompt structure is not complicated, and it can be reused by swapping out the theme and style descriptions.

Sep 14Mon
  1. Cartesia 官方博客22

    TTS models and voices are not the same thing: Cartesia breaks down fixed-speaker, multi-speaker, and voice cloning

    Cartesia's official blog points out that a TTS model and the voice it uses are two separate concepts, with the voice being more like an input than the model itself. Fixed-speaker models receive only text, multi-speaker models can additionally select preset voices, and zero-shot cloning models accept 10 to 60 seconds of reference audio as a prompt; Cartesia's Sonic end-to-end model supports Instant Voice Cloning. Professional voice cloning (PVC) requires fine-tuning the model with a larger recording dataset and can take hours.

Sep 3Thu
Jul 15Wed
  1. Hugging Face Blog36

    Real World VoiceEQ Released: Measuring the Human Quality of Voice AI

    The voice AI evaluation benchmark Real World VoiceEQ has been released, covering 40+ mainstream closed-source and open-source voice models, 15+ evaluation dimensions, and 60+ metrics, spanning ASR, TTS, S2S, and voice understanding. The benchmark is built on over 1 million human ratings, including 785,000 TTS ratings and 48,000 STS ratings. Evaluations show that no system configuration in TTS ranks in the top five across all eight capability groups, indicating that voice models are stronger at "speaking" than at "listening."

Jul 6Mon
  1. Hugging Face Blog22

    PRX Part 4: Our Data Strategy

    The fourth installment of the PRX series reveals its data pipeline: pre-training data is composed of a mix of public and internal datasets, with VLMs used to regenerate long image descriptions, which are then converted into corpora suitable for streaming training. The team switched to the Qwen3-VL text encoder and computes text latents in real time during training, with measured throughput loss of approximately 3–4% (roughly one extra day for a 30-day training run). Images are uniformly encoded at JPEG quality 92, and initial re-encoding is nearly imperceptible in practice.

Jun 9Tue
Feb 3Tue
  1. Hugging Face Blog22

    PRX Text-to-Image Model Training Design: Lessons from Ablation Studies

    The Hugging Face blog publishes the second installment of the PRX text-to-image model training series, using ablation studies to document which training techniques actually work. The baseline is PRX-1.2B with 1.2B parameters, using pure Flow Matching, Flux VAE latent space, 256×256 resolution, a global batch size of 256, and 100k training steps, with data consisting of 1 million synthetic images generated by MidJourneyV6.

Sep 2Tue
Jul 31Thu
  1. Hugging Face Blog30

    Implementing an MCP Server in Python: A Gradio-Based AI Shopping Assistant

    A Hugging Face blog post demonstrates how to build an AI shopping assistant in Python using Gradio's MCP integration, automatically converting Python functions into MCP tools callable by an LLM. The example calls the IDM-VTON virtual try-on model via the Gradio API, combined with VS Code AI Chat and the playwright MCP server to browse web pages, enabling clothing search and try-on effect generation.

Jul 23Wed
  1. Hugging Face Blog29

    TimeScope: How Long of a Video Can Video Large Models Actually Understand?

    The long-video understanding benchmark TimeScope has been open-sourced on Hugging Face. By inserting 5–10 second "needle" segments into videos ranging from 1 minute to 8 hours, it evaluates three capabilities: local retrieval, information synthesis, and fine-grained temporal awareness. Results show that leading vision-language models, including Gemini 2.5-Pro, still struggle significantly with true long-video temporal understanding, and performance degrades more noticeably as video length increases.

  2. 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
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.

Jun 6Thu
Feb 27Tue
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.