AI Film Weekly · Week 42, 2026 · 10.12 — 10.18
AITVHOT
20 major events this week, the most notable being: Black Forest Labs releases the FLUX 3 Image Klein open-weight model; Kuaishou Kling 4.0 fully rolled out, generating videos up to 30 seconds in a single pass; Kling 4.0 officially launched, supporting 30-second single-take shots and multi-frame references.
Black Forest Labs releases FLUX 3 Image Klein open-weight model
Black Forest Labs releases FLUX 3 Image Klein, a 5.6B-parameter open-weight model that supports text-to-image, image editing, and multi-reference generation with up to 10 reference images, and features native 4K generation, pixel-level editing, and bounding-box control.
Model releases/updates
Kuaishou Kling 4.0 fully rolled out, generating videos up to 30 seconds in a single pass
Kuaishou Kling AI announces that Kling 4.0 is officially fully rolled out on klingai.com, generating videos up to 30 seconds in a single pass and supporting multiple extensions up to 2 minutes. The new version improves the stability of large camera movements such as high-speed motion, tracking shots, and orbiting shots, refines character lip-sync matching accuracy, supports 4K and 1080p 10-bit HDR output and 21:9 ultra-wide aspect ratio, and accepts up to 10 images, 5 videos, and 7 subjects per input.
Kling 4.0 officially launched, supporting 30-second single-take shots and multi-frame references
Kling AI officially launches the video model Kling 4.0, supporting direct output up to 30 seconds; previously, in late September, 4.0 Flash early access was opened to some Black Gold annual card and Ultra annual subscribers. The author tested scenarios including city chases, circuses, elevator fights, and figure skating, and believes it has improved in understanding physical laws, complex action transitions, and camera scheduling.
HeyGen releases the HeyGen Voice speech model
HeyGen releases a new speech model, HeyGen Voice, which it calls the first speech model built for video adaptation and which ranks first on the Artificial Analysis speech leaderboard; it is already included in every HeyGen video at no extra charge.
Kling 4.0 hands-on test: how do 30-second direct output, 15 references, and 10 keyframes perform?
Kling 4.0 is fully launched, with a single native generation of up to 30 seconds, support for combining up to 15 reference materials such as images and videos, and input of up to 10 keyframes; the company says it has strengthened stability for high-speed motion and large camera movements such as tracking and orbiting shots.
Kunlun Wanwei's Skywork AI releases Matrix-Game 4.0, running in real time on a single RTX 5090 and open-sourced
Skywork AI, under Kunlun Wanwei, releases Matrix-Game 4.0, which it calls the world's first game world model that can run in real time on a single consumer-grade graphics card, and has open-sourced the model and code. The model leaves game rules to the game engine to maintain and visuals to the video model to generate; when a player is hit, shield durability is deducted before health, and the rules can be rewritten by a Coding Agent using natural language.
Fish Audio releases the open-source speech generation model Drama 3 pre-release version
Fish Audio releases the open-source speech generation model Drama 3 pre-release version, which can convert text into performed speech. The model uses a 35B-A3B hybrid linear attention MoE backbone (initialized from Qwen3.5) and a new codebook-MTP audio decoder; pretraining covers about 80 languages and dialects and hundreds of billions of tokens, and it supports multi-reference voice cloning, natural language instructions, multi-round revisions, multi-character scenes, speech infill editing, and long-audio and streaming synthesis.
Alibaba TaoLive AIGC releases the TaoMate-H3-FL2VA first-and-last-frame control LoRA
The Alibaba TaoLive AIGC team released TaoMate-H3-FL2VA on October 15, a 3-step LoRA that adds first-frame and last-frame control to MiniMax H3's streaming TaoMate-H3 runtime; the team calls it the first known real-time streaming first-and-last-frame-to-audio-video (FL2VA) model.
Krea releases the image editing model Krea 3, at $0.06 per 2K image and most edits completed within 8 seconds
Krea releases the image editing model Krea 3; the company says most edits complete end-to-end within 8 seconds at native 2K, at $0.06 per 2K image and $0.03 per 1K image, with up to 6 reference images per edit.
Tools and Workflows
ComfyUI adds NVIDIA VFX nodes, supporting video super-resolution, frame interpolation, and SDR-to-HDR conversion
ComfyUI turns NVIDIA RTX video effects into custom nodes, including RTX Video Super Resolution, RTX Video Frame Generation, and RTX Video HDR; the models come with the nodes and run locally on RTX GPUs.
Platforms and Regulation
WeChat launches the "Lüpaopao Short Drama" mini program, with all episodes free to watch
Tencent's WeChat quietly launched the "Lüpaopao Short Drama" mini program, whose pages are divided into three main sections: "Free Theater," "Swipe," and "Me." The Free Theater includes categories such as AI dramas, live-action dramas, and animated dramas; all episodes are currently free to watch, with ads inserted between episodes.
WeChat launches the "Lüpaopao Short Drama" mini program, reigniting the short drama battle against Hongguo
The WeChat Short Drama Content Center launched the "Lüpaopao Short Drama" mini program in late September, featuring the slogan "Free good dramas, watch with one tap," with three sections: Free Theater, Swipe, and Me, adopting a joint distribution and advertising model. The company says that after integration, user time spent is 5 times that of the overall WeChat short drama market, and high-quality new dramas with per-user watch time of ≥10 minutes can receive up to 100% revenue incentives. The article compares this with Hongguo Short Drama's scale as of July 2026 of 168 million daily active users and 356 million monthly active users, and notes that WeChat enters with advantages in close-tie social networking and no download requirement, but still has shortcomings in exclusive content and experience.
WeChat launches the "Lüpaopao Short Drama" mini program, benchmarking against Hongguo with joint distribution and advertising incentives
WeChat launched the free short drama mini program "Lüpaopao Short Drama" in late September, produced by the WeChat Short Drama Content Center; there is no standalone app, and the playback page follows the Channels interaction and supports picture-in-picture playback. Content supply follows joint distribution: WeChat does not produce dramas itself but brings content from external short drama mini programs into the official pool and distributes advertising revenue incentives in tiers based on per-user watch time; per-user watch time greater than 10 minutes can receive 100% advertising revenue incentives.
WeChat enters micro short drama incentives, giving producers another quote
WeChat launched a new round of micro short drama mini program incentives starting October 16; high-quality new dramas jointly distributed by Lüpaopao can receive up to 100% revenue incentives, consisting of a 90% cash share and 10% advertising credits, of which the 10% can only be used for ad placement. The article notes that Hongguo's daily active users had reached 168 million in July, and previously producers basically had only one big buyer, Hongguo; WeChat's subsidy gives producers a second quote for the first time, but how long the subsidy can last and whether distribution efficiency can match Hongguo's traffic will determine whether this is a long-term variable or short-term excitement.
WeChat launches the "Lüpaopao Short Drama" mini program to compete with Hongguo for the free short drama market
The WeChat Short Drama Content Center launched the "Lüpaopao Short Drama" mini program in late September, with the slogan "Free good dramas, watch with one tap," featuring three sections: Free Theater, Swipe, and Me; it adopts a joint distribution model and inserts ads between episodes. The company disclosed that after integration, user time spent is 5 times that of the overall WeChat short drama market and retention is more than 2.5 times the market average, and high-quality new dramas under joint distribution with per-user watch time of ≥10 minutes can receive up to 100% revenue incentives.
Industry News
Hengdian joins hands with Volcano Engine on a 235 million yuan AI film and television service platform, while the number of offline film crews drops sharply over the same period
The consortium of Dongyang Deep Space Engine and Volcano Engine won the bid for the Hengdian Film and Television AI Service Platform construction project at 235 million yuan; the tender was launched in July. The project is centered on the Hengdian Film and Television large model and includes 7 deeply enhanced vertical-scenario models, 20 specialized scenario enhancement models, 17 film and television industry professional agents, as well as capabilities such as digital crowd extra generation, actor style transfer, and AI-assisted dubbing, along with a consumer-facing Hengdian Film and Television App.
Character.AI files for bankruptcy, Disney plans to bid $20 million for its assets
Character.AI has filed for Chapter 11 bankruptcy protection in the United States to facilitate a sale of the company and its assets, declaring liabilities of approximately $50 million to $100 million. Disney will act as the main bidder, acquiring some undisclosed assets for $20 million and obtaining a non-exclusive intellectual property license, while providing transition support.
Tips and Opinions
Running MiniMax H3 image-to-video in ComfyUI: a workflow tutorial with synchronized sound
Stable Diffusion Art releases a ComfyUI workflow tutorial for MiniMax H3 (commercial name Hailuo 3.0) image-to-video; a single starting image can output a 24fps multi-shot video in one 5-second generation, while synchronously generating stereo ambient sound, sound effects, music, and lip-synced dialogue.
FLUX 3 Image Klein released: a 5.6B open-source model supporting generation and editing with 10 reference images
FLUX 3 Image Klein is released, a 5.6B-parameter open-weight model that merges text-to-image and image editing into a single model, can accept a text prompt plus up to 10 reference images and output a single image, supports bounding-box editing instructions and in-image text rendering, and supports resolutions up to 4K.
Reverse-engineering Screen Studio with REA: extracting 6 auto-zoom rules and replicating them with claude code in half an hour
The author used the open-source project REA (Reverse Engineer Anything) to break down the Mac screen recording software Screen Studio, generating an evidence map of 3,695 files in 101 seconds, then had claude code write scripts to decrypt more than 60,000 encrypted strings and read out the 6 rules behind auto-zoom.
- 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-W43Hao AI Lab open-sources FastH3 OmniRef, with ComfyUI support on day one
- 2026-W41Google releases image model Nano Banana 2.1, priced at roughly half