Veo 3.1
Google to shut down Veo 3.1 preview models on October 22
The deprecations page points developers to gemini-omni-1.1-flash, Google's August 27 successor, and warns the October 22 date is the earliest possible.
Latest
Everything we've published, newest first.
twelvelabs
TwelveLabs launches Jockey to search and cut video libraries
Jockey becomes the search layer over TwelveLabs' new Marengo 3.5 embedding model and runs inside the ChatGPT and Claude apps.
openai
OpenAI and Meta warn judges deepfakes evade detection
The Advisory Committee on Evidence Rules is weighing a draft Rule 901(c) after a March survey found just 15 of 931 responding judges had faced a deepfake challenge.
Kling
Kling 4.0 opens to everyone with 30-second 4K video
The full model ships 3 to 30 second clips at up to 4K, while Kuaishou still lists 10-bit HDR and two-minute extensions as coming soon.
Theoretically Media
ByteDance's world model re-shoots a movie scene, and Krea 3 lands
Every segment of Theoretically Media's latest episode, checked against the sources it links: a nameless ByteDance world model turning one drawn camera path into a dozen angles inside a Stephen Chow film, Pico's delayed Vision Pro rival, Krea 3's eight-second edits, the report that Sora comes back through a filmmaker startup, an OpenAI engineer charged over a hidden recorder, and two free tools.
deepfake
Colorado deepfake ad of Sen. Cathy Kipp draws free-speech fight
The Independence Institute says the fake video is satire exempt from disclosure; the state elections division says satire must be grounded in humor.
higgsfield
Higgsfield and Luma tie clip cost to how many tries it takes
Both vendors' own tables price a month of usable clips at two attempts each, and both say the retry count moves the bill more than the per-second rate.
sonilo
Sonilo raises $11M to build video-native AI audio
The San Francisco startup, founded by veterans of TikTok's AI and music teams, says its Sound World Model reads footage to generate matching music and sound effects.
Helix World
Helix World generates video and spatial audio in one pass
Noiz AI and HKUST built the world model on the open-source LTX-2.3 model and posted a preview checkpoint, but local runs want an 80 GB GPU.
kandinsky
Kandinsky S2V 1.0 ships 2B and 21B open speech-to-video models
The Kandinsky Lab models animate one reference image at 25 frames per second, and the team's own human study preferred Pro for lip sync over HeyGen and Kling.
OmniHuman
OmniHuman 2.0 generates 5-minute avatar video at $0.07 a second
BytePlus says the model reads tone, emotion and words in the audio and cut the per-second price from $0.12 to $0.07.
Midjourney
Midjourney v8.1 ranks worst of 15 in a VN.ai copyright audit
The startup says Midjourney put copyrighted characters into violent or drug-related scenes in 37% of prompts, and no model came out clean.
FastH3-Omni
FastH3-Omni adds reference images and audio to MiniMax H3
The open-weight model previews on the Reactor API through October 20, when FastVideo says the final weights and training code ship.
standard-chartered
Standard Chartered's F1 campaign film is fully AI-generated
Leo Singapore made the "Move Ahead" hero film with Misideal and Stink, which built a 2026-spec car by hand and rendered it with AI because there was no real car to film.
Marsha Blackburn
Marsha Blackburn's AI Kendrick Lamar video draws mockery
The 12-second clip, posted to promote early voting in Tennessee's governor race, drew a rebuke from Democratic opponent Jerri Green over its use of AI.
Runway
Baseten hosts Runway's video models with a 99.99% uptime SLA
Enterprise customers license Runway's closed weights directly and run Gen-4.5 or Aleph 2.0 single-tenant, though no licensee or price has been named.
Multics
Tencent opens Multics, a world model for 10 players
One 4-player checkpoint generates video for 6, 8 and 10 players without retraining, using communication across player, group and global levels.
Peacock
Peacock puts an AI Andy Cohen in front of every subscriber
Your Bravoverse, live October 13, is a swipeable feed built from 5,000 hours of Bravo, and Wendy's is the first sponsor its avatar names.
Character.AI
Disney bids $20M for Character.AI's code in bankruptcy
Disney Worldwide Services is the stalking horse bidder for a perpetual license to Character.AI's training, safety and serving code and nine patents, not the app.
wavespeedai
AI Cinema Studio turns 18 camera moves into a shot
WaveSpeedAI's new panel writes lighting, lens and a start-to-end camera move into the prompt, while Kling, Runway, Veo and Luma each specify motion differently.