AI Film Weekly · Week 43, 2026 · 10.19 — 10.25
AITVHOT
This week saw a dense wave of updates to video generation models: Tencent Hunyuan open-sourced the world model Hy-Nomas, Hao AI Lab open-sourced FastH3 OmniRef, and ComfyUI launched FLUX 3 Max. On the tooling side, an ecosystem has formed around ComfyUI: Hy-Nomas was ported as a native node, and multiple MiniMax H3 workflows and acceleration LoRAs were open-sourced. At the industry level, two-thirds of overseas vertical dramas in Q3 were AI-generated, and all five major US streaming platforms have completed their short drama lineup.
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.
Model releases/updates
In this sectionOpen weights and hosted launches proceed in parallel: LTX 2.5 supports synchronized audio and native multi-shot, and Kling 4.0 on OnSolo supports single 30-second generation.
Tencent Hunyuan open-sources world model Hy-Nomas, supporting interactive video generation from text and camera trajectories
Tencent Hunyuan has open-sourced its world model Hy-Nomas, which supports free interaction using text prompts and camera trajectories and has long-horizon generation capability with world memory. The model provides lightweight LoRA adaptation for vertical domains such as robotics; after 8-step distillation it can generate 720p@24FPS video on a single RTX 4090. The weights are now available on Hugging Face and GitHub, along with an online Demo.
ComfyUI launches FLUX 3 Max, supporting image-to-video and text-to-video workflows
ComfyUI launched FLUX 3 Max, a latency-optimized version of FLUX 3 Video that can generate 10-second 720/768p clips with native audio in about 12 seconds. The official release also provides links to two workflows, image-to-video and text-to-video, and lists use cases such as comparing multiple candidates, previewing motion while tuning parameters, batch-producing ad variants, and keyframe-guided storyboards.
LTX 2.5 Explained: An Open-Weight Video Model Supporting Synchronized Audio and Native Multi-Shot
LTX 2.5 is an open-weight AI video model that produces audio and video in the same generation pass and supports multiple coherent shots within a single generation. The model is available in standard and distilled versions; weights can be downloaded, called via a hosted API, or deployed on your own GPU. The official team says the new diffusion video decoder improves performance on faces, textures, on-screen text, and fast motion. The article also documents the process of self-hosting on a remote Nvidia GPU and testing lip-sync and facial consistency with dialogue shots.
Kling 4.0 launches on OnSolo
Kling 4.0 is now available on OnSolo, supporting native 30-second video generation in a single pass while maintaining motion stability in fast action and large camera movements. This version offers stereo sound and more precise lip-sync, supports a 21:9 widescreen aspect ratio, and allows up to 15 references and 10 keyframes to be stacked per task.
Tools and Workflows
In this sectionThe ComfyUI ecosystem continues to expand: RMBG adds an interactive canvas and video matting, and CreatorOS Studio open-sources an end-to-end pipeline from storyboard to final cut.
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.
Tencent's video world model Hy-Nomas ported to native ComfyUI nodes, runs on 12GB VRAM
The author ported Tencent's video world model Hy-Nomas to native ComfyUI nodes, with the DiT used as a regular ComfyUI MODEL sampled by the native KSampler, and it runs on just 12GB VRAM.
Stubelius Ultimate H3 0.13 released: chunked generation of MiniMax H3 long videos in ComfyUI, with individual chunks redoable
Stubelius Ultimate H3 0.13 is the author's open-source MiniMax H3 ComfyUI workflow, under the MIT license, made public on GitHub on 2026-10-18.
CreatorOS Studio Open-Sourced: One-Click Storyboard to Final Cut Locally in ComfyUI
CreatorOS Studio is open-sourced under Apache-2.0, providing one-click video creation based on ComfyUI on a local GPU, consuming no credits and bypassing platform content filtering.
taobao-content-ai releases MiniMax H3 Ref2VA acceleration LoRA and ComfyUI workflow
taobao-content-ai released a Ref2VA acceleration LoRA for MiniMax H3, launching 8-step and 4-step PTD versions on October 20, and adding on October 22 an 8-step version trained with PTD plus RefAsymOPD, enabling reference image input with video and audio output.
Platforms and Regulation
Green Bubble vs. Hongguo: Two Distribution Logics for the Second Half of Short Dramas
WeChat's official short drama product Green Bubble launched in mid-October, operated by the WeChat short drama content center, with distribution covering the WeChat search section, the Channels series page, and QQ Browser recommendation slots. The article compares Hongguo's algorithmic closed loop with Green Bubble's social distribution, noting that Hongguo's standalone app has reached 168 million daily active users, while short drama mini-programs within the WeChat ecosystem have 508 million users. The author argues that the two are more likely to grow stronger in their respective arenas, with short dramas moving from one dominant player to duopoly competition.
Industry News
In this sectionShort drama going global has entered a localization phase: the share of locally self-produced dramas rose to 46%, regulation tightened in multiple countries, and AI dramas and comic dramas became market growth drivers.
Owl & Co. report: two-thirds of overseas vertical-drama releases in Q3 were AI-generated
Owl & Co.'s latest "What We Watched on Vertical" report shows that in Q3 2026, two-thirds of the vertical dramas released outside China on 20 leading platforms were AI-generated, compared with 3% in Q1 and 28% in Q2.
HBO Max launches a vertical-screen channel, completing short-drama deployment across all five major US streamers
HBO Max launched the "HBO Max Shorts" vertical short-film channel on July 28, completing vertical short-drama feature deployment across all five major US mainstream streamers: Netflix, Disney+, Peacock, Prime Video, and HBO Max.
Short drama going global hits cultural adaptation: locally produced dramas' share rises to 46%, regulation tightens in multiple countries
MPA data shows the share of overseas locally produced dramas rose from 27% to 46%, while dubbed/translated dramas fell from 73% to 54%, with local dramas achieving paid conversion of 8% to 10%. MPA expects the micro-drama market outside China to reach $3.6 billion in 2026 and $9.5 billion by 2031, of which AI dramas and comic dramas account for about $4 billion.
Tips and Opinions
In this sectionTutorials focus on using MiniMax H3 in ComfyUI, covering reference-to-video, multi-keyframe timelines, and ControlNet motion transfer.
Running MiniMax H3 Fun ControlNet Union video-to-video with synchronized audio in ComfyUI
Stable Diffusion Art released a tutorial introducing how to use MiniMax H3 Fun ControlNet Union in ComfyUI to copy body movements from a driving video while generating 24fps video and stereo audio.
Implementing MiniMax H3 multi-keyframe video and audio generation with a ComfyUI workflow
Stable Diffusion Art released a tutorial on how to use MiniMax H3's Add Guide node in ComfyUI to lock multiple keyframe images, video clips, and audio to specific frame positions on a 24fps timeline, generating a video with stereo dialogue, sound effects, and music in one pass.
MiniMax H3 reference-to-video (ref2va) ComfyUI workflow tutorial
Stable Diffusion Art released a complete ComfyUI workflow tutorial for MiniMax H3 reference-to-video (ref2va), which can combine up to 9 reference images, 3 reference videos and 3 audio clips to generate a 24fps multi-shot video with synchronized stereo dialogue, sound effects and music in one pass.
How to use Seedance 2.5 to turn a single piece of footage into 4 camera angles
Kapwing released a tutorial introducing the use of Claude skills to generate camera instruction prompts, then using Seedance 2.5 in Kai to reshoot a single fixed-camera piece of footage into a multi-camera edit.
Making a 10-minute short drama and vertical short videos with MiniMax H3: how to fix broken hands, abnormal blinking and six legs
The author used the MiniMax H3 Ref2VA model on local ComfyUI to produce a 10-minute comedy short drama, 7 types of vertical short videos and a 60-shot cat character video, at about 2 minutes per shot, or about 3.5 minutes after switching to a 768p 8-step LoRA.
ComfyUI MiniMax H3 timeline tutorial: install the Director plugin to arrange storyboards and generate multi-shot videos
The tutorial explains how to install the MiniMax H3 Director plugin in ComfyUI, drag shots, reference images, and audio onto the timeline, and have the plugin compile in real time the storyboard prompt that H3 reads during training.
- 2026-W46Zhao Liying sues Hongguo Short Drama for portrait rights infringement, seeking 810,000 yuan in damages
- 2026-W45Gemini Omni 1.2 Flash: Faster, more cost-effective, supports 15-second clips.
- 2026-W44ByteDance is embedding short-drama feeds into TikTok's main app, using traffic to rewrite the rules of overseas expansion.
- 2026-W42Black Forest Labs releases FLUX 3 Image Klein open-weight model
- 2026-W41Google releases image model Nano Banana 2.1, priced at roughly half