Skip to content
Date and source

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

Oct 21Wed
  1. 娱乐资本论 · 搜狐公开专栏54

    "Good Girl" hits, Green Bubble enters the fray, and one-person companies rise: the boundaries of the short drama industry are being stretched

    The horizontal-screen medium-length drama "What a Good Girl" landed on Tencent Video with popularity exceeding 20,000, and WeChat renamed its short drama mini program to Green Bubble Short Drama and launched revenue incentives of up to 100%. The scale of 851 million users is not the ceiling; in 2026, the overseas micro short drama market will exceed 6 billion USD, up 67% year on year, and the AI comic drama and one-person company models are changing the industry's profit structure.

  2. Hugging Face Daily Papers47

    Joints, Camera, Action!: jointly generating 3D motion and video for arbitrary skeletons

    Researchers propose a model that can directly jointly generate video and 3D motion for arbitrary skeleton rigs, and estimate the camera to register the motion into the 3D scene. It connects a 3D motion model supporting arbitrary skeletons with a pretrained video model through a small number of cross-attention layers, is trained on about 200K training pairs, and in evaluations drives multiple skeletons by text more faithfully than video-first pipelines, while supporting video-to-skeleton motion, skeleton motion-to-video, and keyframe control.

  3. Hugging Face Daily Papers31

    OverSeer: improving multi-agent world model consistency with a fixed auxiliary camera viewpoint

    OverSeer is a multi-agent world model that, by introducing a spatially fixed auxiliary camera viewpoint beyond the agents' first-person views as a proxy for a world-centric representation, helps the model maintain temporal and cross-agent consistency. The auxiliary view encodes only changes relative to a static reference image via delta-VAE, significantly reducing the tokens and computation required for processing. Experiments on Minecraft, Doom, and real multi-camera captures show improvements in both world and agent consistency.

  4. Hugging Face Daily Papers38

    HRoPE: homography relative position encoding

    HRoPE proposes surface-aware relative position encoding, representing each patch's local geometry as a tilted 3D plane and injecting the plane-induced homography as a relative transformation into attention, to correct perspective distortion and align patch features across views.

  5. Reddit comfyui · 创作与工作流分享54

    Hands-on test and workflow for character replacement using MiniMax H3 reference-to-video plus LoRA

    The author spent about a day testing character replacement using MiniMax H3 reference-to-video with LoRA, and presents unselected early results. The method replaces the person in a video with a character reference image while preserving the original video's motion, scene and audio; the author also lists limitations such as similarity drift, details blending with the original person, and temporal misalignment caused by scene cuts, and says additional prompts help improve consistency.

  6. 霞光社 · 短剧出海与商业趋势74

    China AI Short Drama Overseas Report: Capacity Surge and Traffic Siege, What Are AI Short Dramas Competing On?

    Xiaguang Think Tank released a report on Chinese AI short dramas going overseas, noting that AI short dramas account for 95% of production but only 4% of views, and the industry is shifting from competing on volume to competing on quality. The report shows that the production cost per AI short drama has dropped from 500,000–1.5 million yuan to 50,000–240,000 yuan, with efficiency improving 60–100 times, but overseas traffic acquisition costs account for about 50% of total overseas costs, and the customer acquisition cost per user in North America reaches $5.28. The competitive focus is shifting from content production to traffic acquisition, and compliance has become an entry threshold.

    Why it matters: AI short dramas account for 95% of production but only 4% of views; the report reveals a mismatch between capacity and traffic, providing a reference for overseas players to shift from competing on volume to competing on quality.

  7. 龙爪槐守望者 · 创作者专栏41

    LibTV launches a face-sculpting system to solve the AI video protagonist consistency problem

    LibTV launched a character styling system that uses an approach similar to online game face-sculpting to solve the "same-face" problem of AI video protagonists. The interface is divided on the left by face, body and feature tags, uses linear sliders on the right for single-axis fine-tuning, and offers preset material packs such as eyeshadow and eyebrow shapes below, balancing quick assembly with pixel-level fine-tuning. Customized model assets can be saved with one click as the protagonist of a coherent storyboard.

  8. fal 官方博客28

    EAGLE-4 technical deep dive: Fireworks and fal release speculative decoding benchmark dataset

    Fireworks and fal jointly released a technical analysis of EAGLE-4 and Prompt Enhancement SpecBench, a speculative decoding benchmark dataset for creative tasks. The dataset contains 400 samples, covering 100 each of text-to-video, image-to-video, text-to-image and image-to-image. On Qwen 3.8 27B, EAGLE-4 uses sliding window attention and a low-rank candidate head to reduce draft cost, and the verification tree dynamically adjusts width and depth based on GPU latency measurements.

  9. Loomance · 本地AI动画实测69

    Wan 2.2 comparative evaluation under Loomance and WanGP: Fast Mode and full workflow yield different results.

    The author compared the rendering results of Wan 2.2 under two tools, Loomance and WanGP, and found that the 4-step lightning mode is about 10 times faster but has more errors, while the 30-step full sampling is cleaner. In the six-dog scene, WanGP had 3/5 instances of character melting or merging, while Loomance had 0/5; in high-resolution single-character scenes, the two are close. Choose as needed: use WanGP for rapid changes or low VRAM, use full sampling to maintain character consistency.

    Why it matters: A practical comparison of the same Wan 2.2 model under two local tools offers a reference for creators to choose between speed and fidelity.

  10. ComfyUI Wiki · 工作流与开源创作分享75

    Tencent Hunyuan open-sources interactive world model Hy-Nomas, supporting segmented generation from text and camera trajectories

    Tencent Hunyuan open-sourced the interactive world model Hy-Nomas on October 21, 2026, releasing weights, inference code, and training and data annotation code, along with an online demo and the Hy-Nomas-Bench benchmark.

    Why it matters: Tencent Hunyuan open-sources an interactive world model in which text and camera trajectories can drive long video generation segment by segment, with a benchmark and single-GPU configuration.

  11. ComfyUI Wiki · 工作流与开源创作分享81

    FastVideo releases FastH3 Omni: a reference-conditioned variant of MiniMax H3 that generates video with sound in 8 steps

    FastVideo released FastH3 Omni, a reference-conditioned variant of MiniMax H3; a prompt plus ordered image, video and audio references can generate clips with synchronized sound, requiring only 8 transformer evaluations instead of 50 for base H3.

    Why it matters: FastH3 Omni unifies reference images, video and audio into a single generation pipeline, and provides an item-by-item time comparison with base H3 to help assess local deployment costs.

  12. comfyorg · YouTube 创作视频72

    ComfyUI supports Hao Ai lab's FastH3 OmniRef model

    ComfyUI announced support for Hao Ai lab's FastH3 OmniRef, a model aimed at fast, high-quality video and audio generation. OmniRef enables support for multimodal reference materials, so creators and teams no longer rely solely on text prompts to control generated content; the model can run locally, allowing teams to generate and retain materials on their own hardware without relying on hosted APIs. The official announcement also provides a cloud entry point and a Hugging Face open-weights link.

    Why it matters: ComfyUI integrates FastH3 OmniRef, supporting multimodal reference materials and local operation, so creators can assess building their own video production pipeline accordingly.

  13. Reddit StableDiffusion · 创作者与工作流分享22

    When generating video from reference images with Minimax H3, what to do if a phone or camera always appears in mirror reflections?

    A creator used Minimax H3's reference-image-to-video feature to make an all-POV "Get Ready with Me" influencer-style video. The footage itself looked good, but the person always held a phone or camera in mirror or window reflections, with the problem appearing in about 80% of generations. The creator tried not mentioning the camera, negative prompts, using reference images without a phone or camera, turbo LoRAs such as pmad and dmad, and having the person do other things with their hands, but the phone still appeared and floated around the person; no solution has been found yet.

  14. Reddit comfyui · 创作与工作流分享78

    Hao AI Lab open-sources FastH3 OmniRef, with ComfyUI support on day one

    The FastVideo team at Hao AI Lab has open-sourced FastH3 OmniRef, which generates video with synchronized audio from a prompt plus reference images, short video clips, or sound. The model needs only 8 steps instead of 50, is up to 11.9 times faster than the base MiniMax-H3, has entered the ComfyUI 0.42.2 workflow templates, and has open weights and training code.

    Why it matters: FastH3 OmniRef is released as open weights, produces video in 8 steps, and supports day-one ComfyUI integration, so creators can use it to evaluate local reference-to-video pipelines.

  15. FastVideo 视频模型 · 模型发布80

    FastVideo releases ComfyUI version model weights for FastH3 Omni

    FastVideo released ComfyUI-usable weights for the final version of FastH3 Omni, providing two diffusion models, BF16 and INT8 ConvRot, which can generate video and synchronized audio from prompts and ordered image, video, and audio references.

    Why it matters: FastH3 Omni releases ComfyUI-usable BF16 and INT8 weights and provides the way to connect reference images, video, and audio, making it easier for creators to evaluate local video production pipelines.

  16. Reddit comfyui · 创作与工作流分享78

    ComfyUI-ZeroCopy-Continuity open-sourced: eliminating VRAM reloads and VAE round-trips in multi-shot rendering

    The author open-sourced the ComfyUI custom node suite ComfyUI-ZeroCopy-Continuity, which uses ZeroCopyVRAMPin to keep DiT weights resident in VRAM and ZeroCopyLatentHandoff to pass the previous shot's latent directly on cuda:0, eliminating the VAE decode-then-encode round-trip.

    Why it matters: The author open-sourced a set of ComfyUI nodes that replace VAE round-trips with VRAM residency and direct latent transfer, reducing the transition overhead in multi-shot rendering from 14.2s to 0.4s.

  17. Reddit comfyui · 创作与工作流分享22

    ComfyUI user asks for help: hair shows flickering artifacts after upscaling with the latent upscaler in the H3 workflow OBVPM

    A ComfyUI user, after stitching scenes in the H3 workflow OBVPM, upscaled the video with the built-in h3 latent upscaler, and after an 83-minute render the hair turned into flickering "mirage"-like artifacts; they want to know what this kind of problem is called and how to solve it. The user has a 5060 GPU with 16GB of VRAM and asks whether there is a better upscaling method or alternative workflow than OBVPM.

  18. AI For Real Life58

    Khalil launches AI sci-fi series "The Unmarked" episode 1 and releases a production kit

    Khalil released episode 1 of the original AI sci-fi series "The Unmarked" on the new channel Lazy Motion Pictures, telling the story of a mother and daughter facing a system threat after an identity-marking ceremony goes wrong. He also launched a production kit for episode 1, including a 171-page production guide, 74 prompt entries, a clip index, and edit trim information, and says generating beautiful shots is only a small part; the difficulty lies in making shots carry the narrative functions of performance, dialogue, emotion, camera placement, continuity, editing, and sound.

  19. Curious Refuge film gallery22

    V-Andrei's AI short film "You Tied It Wrong": a samurai goes to a rendezvous after battle to bring his niece home

    The AI samurai short film "You Tied It Wrong," created by V-Andrei, tells of a masked samurai who climbs toward a garden after battle to fulfill a promise to his brother—to bring his daughter home. Fragmented memories, a desperate confrontation, and the recurring image of a loose knot form a somber story about family, failure, and fragile hope.

  20. Curious Refuge film gallery22

    WRAI's AI short film "Lost & Found": a 1970s train station lost-and-found office

    WRAI's AI short film "Lost & Found" is set in the lost-and-found office of a 1970s train station, where a seasoned clerk receives visitors looking for or leaving items among towering shelves of numbered objects. The nostalgic setting and quiet, thought-provoking premise turn everyday lost-item handovers into a reflection on loss, discovery, and the answers people seek.

  21. Curious Refuge film gallery22

    "Metamorphosis" S1 Ep.04: AI sci-fi short film "Metamorphosis" Episode 4

    Episode 4 of the AI sci-fi short film "Metamorphosis" by Rafael Taveira Pedroso and Samantha Shields tells of Subject 01 (Anya) being transported to an unregistered desert area, where the mysterious guardian Gael awaits her. Set in 2375, the story shifts from an auction that sells emotions as upgrades to an ambush in a sandstorm, presenting the commodification of sentient life and the possibility of resistance.

  22. Reddit comfyui · 创作与工作流分享78

    ComfyUI-RMBG releases v3.3.0, adding interactive canvas compositing and VRMBG video matting nodes

    ComfyUI-RMBG releases v3.3.0, adding a Canvas interactive multi-layer compositing node, temporal video matting based on BRIA AI VRMBG-3.0, FeyNoBg fine matting for hair strands and glass, and YOLO26-seg and YOLO11-seg segmentation support.

    Why it matters: This update integrates matting, detection, and compositing into the same set of ComfyUI nodes and adds temporal video matting, allowing evaluation of replacement costs against existing production workflows.

  23. GitHub:AI视频开源项目发现36

    Qianshuixia open-sources shrimpreel: a SaaS template for generating viral-template AI videos from a single photo

    Developer Qianshuixia (qianshuixia) has open-sourced shrimpreel, a SaaS site template for "generating viral-template follow-shot AI videos from a single photo," where each trending keyword can be replicated into an independent site that goes live after changing the configuration and assets.

  24. Reddit comfyui · 创作与工作流分享71

    How I built an end-to-end AI face-swap video workflow

    The author breaks down an AI face-swap video workflow from raw footage to finished video, using ten folders numbered 00 to 09 to divide stages such as shot planning, reference images, generation packages, frame alignment, final compositing, and quality control. In the generation stage, the pipeline is run twice on ComfyUI with three NVIDIA DGX Sparks, first a 1280×736 draft and then a 1920×1088 refinement; an 8-second 1080p-level clip takes about 1 hour.

    Why it matters: The author reviews the full workflow of a face-swap video with 54 shots and 8163 frames, focusing on easily overlooked engineering steps such as frame alignment and version management.

  25. Reddit StableDiffusion · 创作者与工作流分享62

    CastWorkshop open-sourced: a character and scene studio based on ComfyUI and Ollama

    CastWorkshop has been open-sourced and is positioned as a coordination layer between Ollama and ComfyUI, allowing users to create persistent characters in natural language, place characters in the same scene, and generate images and videos. It uses Krea 2 and MiniMax H3 by default, also supports connecting to existing ComfyUI workflows, and recommends a hardware baseline of 16GB VRAM and 64GB RAM. Other features include character personality and dialogue history, two-character scenes, video continuation, and Telegram integration for mobile use.

Oct 20Tue
  1. TelosCompass · note AI视频创作实践38

    Generating PR video storyboards with ChatGPT: revising the AI plan scene by scene

    When producing a 15–30 second PR video, first submit the subject, audience, core value, and format conditions to ChatGPT all at once to generate a storyboard plan, then revise each scene by rewriting atmospheric words like "soft rainy scene" into "what is seen, what moves." In the example, a 20-second video is split into four 5-second scenes, with the protagonist fixed as a dark blue long umbrella, and people, dialogue, and hand movements all removed; the final scene reuses the previous scene's footage with text added. Before use, check the four items of purpose, facts, production, and duration, and keep only one version of the planning memo.