Jimeng
How ByteDance's Jimeng Short Drama Studio works
The one-workspace tool runs on three in-house models and can regenerate a finished drama's visuals for another market.
Explainers
How it works and what it means.
MiniMax H3
MiniMax H3's 2K finish is now an open 3-step pass
The refiner and decoder still carry MiniMax's H3 license, which excludes the United States, the European Union, the United Kingdom and South Korea.
synthetic performer
New York and California require synthetic performer ad disclosures
New York's law is already in force; California's SB 1050 takes effect January 1, 2027 and prescribes the words the label must use.
ComfyUI
ComfyUI 0.43 adds Kling 4.0 as a paid video API node
The update routes Kling 4.0, PixVerse V6 Max, Mirelo SFX and HeyGen voice through paid Comfy API nodes, while Fibo 2.0 and MOSS-Image edits stay on your GPU.
iOS
App Privacy Report: 542 events, and the number a viral post got wrong
iOS's hidden log counts how often apps touch your mic, camera, photos and contacts, and which servers they call. Two posts read 48 hours of it with Claude Code. The log is real; one headline figure is not.
korea
Korea's AI film boom is outrunning its copyright rules
Generative AI has pulled Korean AI films close to live-action quality, but its copyright, likeness and performer rules remain unsettled.
ByteDance
ByteDance opened a micro-drama factory, and here is how to use it
The studio lives inside Jimeng, which is China-only today, but the pipeline it automates and the models underneath it are reachable from outside through Dreamina and the Seedance 2.5 API. A step-by-step guide to making a 60-episode vertical drama, from premise to cost sheet.
AI face swap
AI face swaps: what to do when your clip is reskinned
A UK car creator found her drag-strip video reposted with a stranger's face, and Instagram told a reader the copy broke none of its rules.
MiniMax H3
MiniMax H3 cuts sampling to 4 steps with two open releases
FastH3 Omni ships as an 8-step ComfyUI student; Taobao's PTD LoRAs reach 4 steps, and neither release has independent quality benchmarks.
filmflow
FilmFlow turns a text prompt into a Blender camera path
The Shanghai Jiao Tong University-led system ships with CinemaBench, 150 Blender scenes and 750 camera instructions, but no code is released yet.
AI likenesses
AI likeness deals: who signs them and what the rules allow
Sunny Hostin and the cast of Higgsfield's Cully Hill Boys licensed their likenesses; Catherine Zeta-Jones prefers real performers and Nicolas Cage refused an AI waiver.
nvidia
Nvidia's $4,999 DGX Spark runs some AI video models, not all
NVIDIA's six OEM partners start selling the 64GB box on October 23, and two units linked by a QSFP cable pool to 128GB for models up to 200 billion parameters.
comfyui
ComfyUI adds VEED nodes to lip-sync a still photo
The three cloud nodes also re-sync a speaker's mouth in footage you already shot and cut a subject out with an alpha matte.
verus-insight
Kling and Peking University publish a detector that shows its work
A Peking University and Kling team's 8B model lifts detection accuracy, but its own benchmark shows it still misses most marked artifacts.
google
AI video's preview trap keeps retiring models mid-project
The deprecation table names gemini-omni-1.1-flash as the replacement, and the same date retires Gemini Omni Flash's own preview build.
SlopTV Games
SlopTV Games review: Cold-Gas Teslas and a 1984 NFL rebuild
A browser arcade with 35 titles, no install and no sign-up, and four in-house games worth the trip: an eight-stage Cold-Gas Tesla shooter, a top-down cocaine co-op thriller, a six-fighter political parody, and a 4.2 GB rebuild of ESPN NFL 2K5 set in the 1984 season.
Kling 4.0
Omni reference: what AI video models now take as input
Kling 4.0 takes up to 15 reference assets in one generation, while MiniMax H3 and the open FastH3 Omni cap a request at 12 files.
deepfakes
How to tell if a video is AI-generated, and where the checks fail
Provenance data, detector models and a court's reliability test each catch different fakes, and the newest research says the pixels alone are getting harder to read.
Higgsfield
Higgsfield CLI generates AI images and video from a terminal
The open-source tool signs in through a browser instead of an API key, and charges generations to the same plan credits as the web app.
artificial-analysis
Artificial Analysis re-based its image-to-video leaderboard
The board now ranks audio-enabled models as AA-Video-I2V v1.1, its Elo values no longer carry over, and Vidu Q4 Preview is missing from it.