Skip to content
Date and source

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

Oct 16Fri
  1. Reddit comfyui · 创作与工作流分享18

    What is currently the best workflow for extending H3 video clips?

    A user asked about the current best workflow for extending H3 clips while minimizing quality degradation, and mentioned the Harley Quinn examples commonly seen in the community, but does not know how to generate clips several minutes long like theirs.

  2. Reddit comfyui · 创作与工作流分享10

    Can 16GB RAM + RX 6700XT 12GB run video generation? ComfyUI user asks for help

    A ComfyUI user posted on Reddit asking whether their setup—RX 6700XT 12GB, Ryzen 5 3600, 16GB RAM, 1TB SSD + 2TB HDD—can run video generation. The user currently generates images in about 1 minute with the Flux schnell quantized model and about 20 seconds with Juggernaut XL, and wants to know whether a new PC is needed.

  3. Reddit comfyui · 创作与工作流分享12

    How to find style LoRAs compatible with character LoRAs? ComfyUI user seeks ZIT workflow experience

    A ComfyUI user found that when mixing style LoRAs with character LoRA on Civitai, some combinations work well while others are completely incompatible. The user employs ZIT and hopes to find a more efficient method than 'download and test one by one' to screen for compatible style LoRAs, and is soliciting tips or indicators for compatibility.

  4. Reddit StableDiffusion · 创作者与工作流分享10

    Seeking lightweight local AI models on an M1 MacBook Air (8GB unified memory): single-image + audio-driven character animation

    A user on Reddit is seeking an AI model that can run locally on an M1 MacBook Air (8GB unified memory) to drive realistic character animation from a single reference image, audio, and prompts for series production. Their performance targets are a minimum of 480p, preferably 720p, 24 FPS, segments of 8 seconds or more, with generation time under 1 minute per segment (ideally within 30 seconds), and support for clip extension similar to Google Flow.

Oct 15Thu
  1. Reddit comfyui · 创作与工作流分享10

    How can ComfyUI beginners do video character replacement and motion control? Seeking an introductory tutorial

    A ComfyUI beginner wants to use a local deployment to achieve video character replacement and motion control for future exploration of UGC content, but cannot find a suitable introductory tutorial online; existing videos mostly target intermediate users and have a high terminology barrier. The ready-made workflow they downloaded does not run well, with characters blending together, and they hope to reproduce the effect of Wan 3.0 on the Higgsfield website, seeking a usable workflow or tutorial.

  2. 歸藏8

    Feature Request Feedback for Grok Bot Avatar Switching

    @poteto I'd like my Grok Bot to be able to switch avatars on its own, but I've run into two issues: The Bot cannot change its avatar animation for different states (idle, working, etc.). My main Bot cannot see what avatar other Bots are currently using. I hope both of these features can be supported!

  3. Reddit StableDiffusion · 创作者与工作流分享12

    How to train a LoRA with GPT Image 2?

    A creator is trying to generate 15,000 images with GPT Image 2 to train a LoRA, but has run into the problem that the Batch API file size limit is too small, while making requests one by one is too costly and easily triggers rate limiting. Currently both prompts and images are processed through the API; the author is seeking a solution and stresses that the training goal is complex details rather than mere style.

  4. Reddit StableDiffusion · 创作者与工作流分享8

    What jobs can you find with Stable Diffusion skills?

    A software engineer with years of mobile development experience spent the past year going deep into Stable Diffusion, writing their own ComfyUI nodes and Python scripts, automating workflows, and doing LoRA training and open-source model fine-tuning. After losing their job last month, they want to turn this hobby into a career. They feel they are not yet qualified to call themselves an AI engineer, have limited PyTorch experience, and consider full-stack software engineering their strength; they ask which positions to apply for and what to learn next.

  5. Reddit StableDiffusion · 创作者与工作流分享12

    Flux 2 and 3 always add an extra key light during generation; user asks how to remove unwanted white light

    A user asked for help on Reddit, saying that when first trying Flux 2 and 3, no matter how the prompt was written or how the environment changed, the model would add an extra unprompted whitish key light to the image, while they only wanted red and blue lighting. They tried using a reference image to maintain character consistency and suspect Flux carried over the lighting from the reference image, but could not resolve it.

  6. Reddit StableDiffusion · 创作者与工作流分享12

    Would You Watch Me Make an AI Video from Scratch Using MiniMax + Claude Opus 5.5?

    A creator posted on Reddit asking if anyone would like to watch a live stream of them making an AI video from scratch using MiniMax and Claude Opus 5.5. They plan to fully showcase prompts, experiments, mistakes, and the real time it takes to bring an idea to life — no shortcuts, just the authentic creative process — and are collecting interest in the comments.

Oct 14Wed
  1. note · miyuu_2026 · 图像视频创作分享15

    ComfyUI poolside girl generation model comparison (English prompts)

    A ComfyUI poolside girl generation comparison test published on 10/14, using English prompts to compare image outputs across multiple models. The original body text contains only an invitation to vote on character popularity and a "continue reading" prompt, and does not disclose specific model names, parameters, or scoring data.

  2. SuSu_酥酥38

    Claude Fable 5.5 appears in the Azure catalog

    🚨Alert! Claude Fable 5.5 has appeared in the Azure East US 2 catalog with the status "GenerallyAvailable"! The LiteLLM configuration has been synced: 1M context / 128K output / $10/$50 pricing, and its capabilities will surpass the current strongest Opus 5.5. If it were released today, would it immediately set a new SOTA in multiple domains? #Claude #Anthropic #Fable55 #AI #Azure #LLM #Dario

  3. Reddit StableDiffusion · 创作者与工作流分享10

    How can motion control videos made with MiniMax H3 look more realistic?

    A user asked for help on Reddit, saying that when making motion control videos with MiniMax H3, they could never achieve the desired realism and the textures looked "plastic"; they were unsure whether the prompt was written incorrectly or the reference image was chosen poorly. The post gave no specific solution or parameter settings.

  4. Reddit StableDiffusion · 创作者与工作流分享8

    How can I generate natural, realistic AI videos locally with Forge Neo?

    A creator who can already generate natural, realistic images locally is asking how to take the next step and turn characters into short videos, and admits to knowing nothing about video generation. They want to know the principles of video generation, what tools to use, whether it can be done with checkpoints on Forge Neo, and whether LoRAs work with it as they do for images and whether character style can be kept consistent.

  5. 歸藏10

    歸藏 solicits unpublished skill topic suggestions

    歸藏 says he still has many skills unpublished and is asking netizens which one they'd most like to see first. Candidates include spoken-video subtitle generation and editing, video packaging, WeChat official account formatting, HTML document formatting, and a half-finished Personal Agent personal context mounting tool.

  6. note · lofty_eagle5366 · 图像视频创作分享4

    The limit of "cute" in AI illustration: a quiet creative experiment

    The author created AI illustrations on the theme of "cute," attempting to quietly express the limit of what they consider "cute." The original text only gives the theme word "cute" and the keyword "儚" (fleeting, ephemeral), and does not disclose the model, tools, or specific works used.

  7. 数英8

    BU蓝昊 (Guangzhou) hiring post-production (AI direction)

    BU Lan Hao (Guangzhou) is hiring for post-production (AI direction), requiring 2+ years of work experience, an art background, and the ability to independently complete the full post-production workflow for TVC advertisements, including editing, compositing, and animation, while leveraging AI technology to produce complex commercial TVC works. The position requires proficiency in AI tools such as Jimeng, Kling, tapnow, and ChatGPT, mastery of post-production software such as Adobe, and a portfolio of mature commercial TVC works.

Oct 13Tue
  1. 开源中国15

    4th Open Atom Competition 'Flight to the Earth's Core' AIGC Creation Track Calls for 2-5 Minute Ultra-High-Definition Videos, Total Prize Pool 330,000

    The 4th Open Atom Competition · 2026 Open-Source Industry Solution Innovation Competition (Beijing Open-Source Innovation Competition) Track 3 'Flight to the Earth's Core' AIGC Creation Track has opened for submissions, targeting cultural tourism scenarios such as immersive flying cinemas, calling for 2-5 minute ultra-high-definition videos with themes centered on the underground world. This track offers 16 award slots with a total prize pool of 330,000 yuan.

  2. 雷峰网38

    XGRIDS Builds the World's Largest Air-Ground Integrated 3D Gaussian Model 'Minning Town' with 6.2 Billion Gaussians and 15 Billion Point Clouds

    XGRIDS used 186,000 aerial and ground images and 15 billion radar point clouds to train a continuous 3D Gaussian model called 'Digital Minning Town,' covering 23.5 square kilometers and containing 6.2 billion Gaussian ellipsoids, with reconstruction error better than 2cm. The model was collected by 3 field personnel over 7 days, and was brought into Unreal Engine via the LCC Plugin for UE for real-scene reconstruction and visualization in CCTV's documentary 'Minning Avenue.'

  3. Animation Magazine15

    "Gnome Alone 2" to Premiere at the American Film Market

    "Gnome Alone 2: Game of Gnomes" will premiere at the American Film Market (AFM); it is the sequel to Peter Lepeniotis's 2017 3D CG family film "Gnome Alone." The film is produced by SC Films International and is the first project under the company's multi-year, multi-film agreement with AI-native animation studio SiXiTZ Studios.

Oct 12Mon
  1. 昆仑万维 · SkyReels8

    Kunlun Wanwei SkyReels releases Matrix-Game-4.0

    Kunlun Wanwei SkyReels launched Matrix-Game-4.0. The model is currently not deployed by any Inference Provider and does not yet support online inference calls. The README on its model page is empty, and the platform does not track download counts for the model.