Skip to content
Date and source

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

Oct 23Fri
  1. Hugging Face Daily Papers31

    FAPL: making MLLM latent variables truly drive cinematic video generation

    Researchers propose Factor-Aware Predictive Learning (FAPL), which, through grounded training of a latent variable generator with a frozen MLLM and connector, enables the Video DiT to truly utilize MLLM latent variables rather than bypassing the native conditioning pathway. The method combines factor-span replacement to achieve factor-aware control over characters, scenes, actions and camera, and outperforms baselines on all three cinematic video generation tasks: sparse input to video, factor-aware editing and multi-shot generation.

  2. 四川省广电局 · 行业动态22

    The 4th "Audio-Visual China" Forum held in Beijing, Sichuan cultural micro-dramas push further into overseas markets

    On October 22, the 4th "Audio-Visual China" Online Audio-Visual Academic Frontier Forum and the Dialogue on Overseas Dissemination and Value Co-creation of Micro-Dramas in the AIGC Context was held in Beijing. the Audio-Visual Art Research Center of Communication University of China released the research report "2026 Micro-Drama Creation and Overseas Dissemination—A Case Study of Sichuan Regional Culture 'Going Global'". At the meeting, multiple parties jointly launched the co-construction of the "Sichuan Culture Going Global Micro-Drama Script Library" and marked the conclusion of the "Global Celebrities Bashu Tour" event, which achieved total exposure of over 270 million views across the internet.

  3. 钛媒体58

    AI 3D "top three" raise over 8 billion yuan in half a year, with VAST, Meshy and Yingmou each taking different commercialization paths

    The three AI 3D companies VAST, Meshy and Yingmou raised a combined total of over 8 billion yuan in half a year. VAST has raised over 5 billion yuan cumulatively at a valuation of $2 billion; Meshy completed $400 million in funding at a valuation of $1.5 billion; and Yingmou completed a Series B of several hundred million yuan in June.

  4. Reddit comfyui · 创作与工作流分享22

    Krea2 edit node loses detail: user seeks a way to preserve source image details 1:1

    A user reports that although Krea2 is not an editing model, operating through the edit node causes the output to lose a large amount of detail and exhibit pixel drift. They have tried various sampler/scheduler combinations, 8/10/12 steps, secondary upscaling, Krea2 enhance LoRA, raw/turbo models and the Wan2.1 VAE, all without noticeable improvement, and ask whether they can only wait for the local version of Krea3.

  5. Reddit comfyui · 创作与工作流分享22

    What is the best workflow in ComfyUI for maintaining model consistency across clothing and backgrounds?

    A user asks how to use dozens of photos of the same model in ComfyUI to generate fashion images with consistent facial and body proportions while changing clothing, background, pose, camera angle and lighting. Their core question is whether to train a character LoRA, use reference-image conditioning, or combine both, and they hope for model and workflow recommendations as well as advice on preparing reference photos.

  6. Luma News60

    Luma integrates Kling 4, focusing on fast action and large camera movements

    Luma announced that Kling 4.0 has been integrated into the platform, available to all Luma customers and selectable directly within the Luma board. The model focuses on fast action and camera control, supporting tracking, orbiting, push-in, pan, and shot-size changes, with character expressions closer to the description and on-screen text remaining stable during camera movement. It supports text-to-video and image-to-video, and offers four aspect ratios: 16:9, 9:16, 1:1, and 21:9.

    Why it matters: Luma integrates Kling 4.0, allowing creators to compare camera movement and action performance within the same project board before deciding on shot allocation.

  7. Variety 产业新闻18

    Peter Chan rejects AI-generated content, says two-hour theatrical films will not be replaced by vertical short dramas

    At the Singapore International Film Festival, Peter Chan said he might use AI as a technical tool, but "absolutely will not use AIGC," and believes two-hour theatrical films will still hold their ground at least in his lifetime. He said CGI has long relied on AI for years, just without calling it that, but "they don't understand art; films are still made by people." His new film "The Road Not Taken" focuses on Alzheimer's disease and a broken family, and has already premiered in Toronto.

  8. Krea 官方博客58

    Krea launches multi-user real-time collaboration for the Nodes workflow editor

    Krea announced that Nodes supports multi-user real-time collaboration, allowing team members to simultaneously open the same project and edit the same canvas, with changes, cursors, selection states, and run results synced in real time. Edit sharing and real-time online status are available only to Krea Enterprise; view sharing (workspace view, specified-person view, public link) is available on all plans; run costs are borne by the initiator in their workspace, undo only reverts one's own changes, and version snapshots are taken at most once every five minutes and retained for five days.

  9. 短剧深水区 · 澎湃公开专栏76

    Do the math: how much can you earn from collecting coins for 1,176 consecutive days on Hongguo Short Drama?

    Hongguo Short Drama announced a phased discontinuation of its coin feature. Collecting coins with full attendance for 1,176 days yields a total return of roughly a few hundred yuan to over 1,700 yuan. Starting November 1, it will no longer be offered to new users, and by December 31 it will be fully discontinued for all users, with existing benefits retained. Hongguo also launched Xiaoguo AI, powered by Doubao's large model, shifting from subsidized user acquisition to content and technology distribution.

    Why it matters: Hongguo discontinues its coin feature and launches Xiaoguo AI; the short drama industry is shifting from subsidized user acquisition to content and technology distribution, and creators need to shift to competing on content quality.

  10. Reddit StableDiffusion · 创作者与工作流分享15

    How to add multiple LoRAs to the Minimax H3 default t2v workflow

    A user on Reddit asked how to add additional LoRA nodes to the default ComfyUI text-to-video workflow of Minimax H3. The default workflow comes with a turbo LoRA and multiple switch nodes; the user said they can already configure multi-GPU operation but cannot understand the switch logic, and LoRA is simply ignored during actual runs.

  11. 漫剧出海 · 每日观察70

    Vertical short dramas shift from a China-going-global playbook to platform infrastructure

    Vertical short dramas are shifting from an experimental export from China to default infrastructure built by major platforms. TikTok has brought publishing, viewing, payment, and advertising in-app, while Hollywood studios have simultaneously announced they are entering vertical drama production. On TikTok's side, Minis Center handles viewing unlocks, Drama Center handles content-provider publishing, and Growth Max handles automated advertising; in the first half of 2026, short drama revenue sharing totaled $82 million.

    Why it matters: Sorting out TikTok's in-app short drama loop and the signals of Hollywood's simultaneous entry, to help creators judge shifts in the focus of distribution paths and monetization models.

  12. 漫剧出海 · 每日观察58

    ReelShort parent Crazy Maple signs with Orange, bringing short dramas into carrier phone plans

    On October 15, ReelShort parent Crazy Maple Studio signed a multi-year cooperation agreement with African and Middle Eastern telecom operator Orange to distribute short dramas to 17 markets through the Max it super app, initially reaching over 180 million subscribers; users who purchase a plan including data can watch the library for free during the validity period.

  13. z-ab · 开源短剧与创作项目分享12

    October 23 AI Tech Daily: OpenAI prepares $30 billion funding, GPT-6 Luna decision API opens, and 23 other items

    OpenAI is advancing a new $30 billion funding round, with a post-money valuation expected at $1.25 trillion, and has fully opened a decision API based on GPT-6 Luna at $0.10 per million input tokens, with response speed 10 times faster than regular calls. NVIDIA open-sourced OpenShell, providing a secure private runtime for large-scale autonomous AI agent clusters. Fotor launched the Fotor Agent video production tool covering everything from AI generation to final editing.

  14. note · tidaa · 图像视频创作分享38

    Midjourney prompt generator ver.6.1 released: camera settings return, new prompt format toggle added

    Midjourney prompt generator ver.6.1 released, restoring detailed camera settings and adding a prompt format toggle. Users can separately specify camera model, lens, ISO, aperture, and shutter speed, or use preset mode to select a shooting style with one click; output format can be switched between natural language and structured lists. Pose presets increased from 8 to 12, adding four new scenes: Cozy Reading, Resort Holiday, Phone Break, and Smart Look.

  15. 虎嗅 · AI影视与短剧产业58

    WeChat's Green Bubble enters short dramas, Hongguo takes coin feature offline and tightens revenue-sharing rules

    New entrant Green Bubble Short Drama in the WeChat ecosystem has not yet shaken Hongguo Short Drama, which has 168 million daily active users; Hongguo announced it will phase out the coin feature, no longer opening it to new users from November 1 and fully taking it offline by the end of December. As AI short drama production capacity surges, platforms are tightening revenue-sharing and cooperation rules; in September 2026, the number of new native AI dramas/comic dramas on Douyin reached 51,800, with only a 0.72% rate of exceeding 100 million views, and screenwriter fees have fallen accordingly.

  16. Krea 官方博客62

    How Krea uses Krea: from a layered tech stack to a self-trained agent team.

    The Krea team used its own tools to produce the Krea 3 launch video and benchmark video, cutting costs from over $200,000 to about $300 and time from 2 months to within 24 hours. The workflow began with Figma storyboards, went through Krea MCP, Claude, and After Effects plugins, and was ultimately completed by the self-trained agent Wobble, which produced 116 versions of the benchmark video.

    Why it matters: The Krea team has published a layered workflow from Figma storyboards to self-trained agents, along with prompts and feedback mechanisms, so creators can adapt this approach.

  17. Reddit comfyui · 创作与工作流分享67

    Author tests AI character replacement with MiniMax H3 in ComfyUI and releases a workflow tutorial

    The author tried MiniMax H3's Reference-to-Video workflow in ComfyUI, using AI-generated characters to replace people in the original video while preserving the original footage's action and performance, and also attempted to replace the background environment.

    Why it matters: The author tested MiniMax H3's reference-to-video process locally, organizing character replacement and motion transfer into downloadable workflows and assets.

  18. Wireflow · AI短剧与视频制作教程35

    JSON2Video 2026 pricing: plans and real cost per minute

    JSON2Video pricing is $199.50 per year for Hobby and $49.95 per month for Professional; 1 credit corresponds to 1 second of video at any resolution, and failed renders are not billed. Calculated per minute, Startup is lowest at $0.20, 20% lower than Professional and about 40% lower than Hobby; one-time credit packs are more expensive but do not expire. Note that subscription credits reset monthly, a single video can be up to 30 minutes long, and rendered files are retained for only 7 days.

  19. Reddit StableDiffusion · 创作者与工作流分享20

    Seeking help with MOSS-style LoRA training: 150 images, 30 epochs, and loss stuck at 0.44

    A beginner trained a MOSS-style LoRA with FizGig's default settings, carefully selecting and strictly captioning 150 images and setting 30 epochs, but by epoch 18 the loss could not go below 0.44 and kept rising back between 0.48 and 0.56, so they stopped training and asked the community whether to continue or adjust parameters.

  20. note · ktcrs1107 · 图像视频创作分享15

    Making an otherworld hero into a video with Wery AI, and he only knows how to exchange business cards | ハロウィンアート2026

    A creator used Wery AI to make two videos about "otherworld hero Takeshi": one about an ordinary office worker summoned to another world while exchanging business cards at a trade show, and another about him continuing to hand out business cards in an otherworld tavern. The character's merchandise sold for only 300 yen, the demon lord subjugation made no progress, and the author also tried Wery AI's Sona love fortune-telling and AI presentation generation features.

  21. ComfyUI Wiki · 工作流与开源创作分享72

    MOSS-Image's first instruction-editing LoRA released, with native support merged into ComfyUI the same day

    On October 22, 2026, MOSS-Image got its first instruction-editing LoRA, chinoll/Moss-Image-Edit, a rank-64 adapter that takes 1 to 4 reference images and a text instruction and covers changes to color, clothing, lighting, season, and visual style.

    Why it matters: The community adds an instruction-editing LoRA for MOSS-Image; the article gives two integration paths—ComfyUI native and a node pack—along with default parameters, which can be evaluated against existing editing workflows.

  22. Reddit comfyui · 创作与工作流分享15

    How do you build a realistic AI shoot workflow with complex equipment for a fiber-optic network company?

    A web designer asked for help in the ComfyUI community on Reddit: making realistic AI photos for a fiber-optic network infrastructure company, requiring the same group of workers across different scenes while preserving specific technical equipment; even with reference images, it was hard to reproduce equipment details. They think fashion shoots or people holding products are far easier than this kind of niche scenario and hope to find a YouTube tutorial or workflow suggestions that fully demonstrate a consistent AI shoot process.

  23. Reddit comfyui · 创作与工作流分享15

    With a Ryzen AI Max 395 and 64GB of memory, how much VRAM should be allocated to the GPU?

    A user running ComfyUI with MiniMax models on a machine with a Ryzen AI Max 395 and 64GB of memory found that after allocating 16GB to the GPU, the logs showed a lot of offloading, so they set VRAM to 48GB in the BIOS, after which ComfyUI started crashing. The user asks whether VRAM allocation affects performance under a unified memory architecture.

  24. Reddit comfyui · 创作与工作流分享12

    Seeking open-source alternatives to Nano Banana: i5-10600K + Arc B580 configuration

    A user posted in the comfyui subreddit asking whether there are open-source alternatives to Nano Banana and hoping the community would recommend some. The user said they are not after Nano Banana-level capability but do not want to pay monthly and want less censorship; their setup is an Intel i5-10600K, 32GB of memory, and an Intel Arc B580 graphics card with 12GB of VRAM.

  25. Reddit StableDiffusion · 创作者与工作流分享20

    How do you avoid flicker or colored dots when extending video with Minimax H3 in ComfyUI?

    When extending video with Minimax H3 in ComfyUI, the user saw flicker at the start of the second segment and colored dots throughout the whole segment; they tried adjusting context frames, window 39/90/141, snap cuts, quality, and aspect ratio, and switched between the ref2va and fl2va models, all to no effect.