Last week in AI video, Oct 12 to 18:
Kling 4.0 is now fully available worldwide, and two new tools can score your video with music and sound effects.
Here's what video, image and audio creators should know
5/ Radar and quick hits
MOSS Image: 5B open weight image model that places text and objects with JSON boxes. License unconfirmed.
OmniHuman 2.0: talking video from one image and a voice clip, via API.
Drama 3: open weight TTS you direct with bracket tags.
The script is based on public reporting as of Sep, Oct 2026 and simplified for a general audience, so some nuance is bound to be missing.
Shipping, logistics, and Arctic folks: what did I get right, and what would you push back on?
Always happy to chat
The Arctic route has been all over Korean news lately. Ten years ago, I crossed those same waters as a deck officer, so this one felt personal.
So I made a 90-second motion graphics video explaining its current state.
Quick hits HyperFrames Studio: agent built video drafts you edit by hand (Mac, Linux)
Claude Motion: prompt to editable MP4 explainers (Team and Enterprise beta)
SynthID Detector: checks AI watermarks from Google, OpenAI, NVIDIA,
Weekly update for AI Video Creators
(~10/11)
Open source AI video now ships with sound, and AI just moved inside After Effects.
5 tools AI video creators can use this week, plus 3 to watch
On the radar Tencent Prism: 2K video with audio, preview weights, needs an 80GB class GPU
Midjourney Thinking Mode: a rerun that rethinks your prompt, Alpha test only
ArtCraft: 7 open source creative apps, still in alpha
7/ The shift this week: from regenerating everything to fixing just the part you need.
Which camp are you in?
A) Waiting for the full Kling 4.0
B) Moving to Wan 3.0 now
Reply and tell me why.
Most AI news last week (~Oct 4)
was about chatbots and agents.
If you make AI video, here's what actually changed for you:
Kling 4.0 opened early access
Wan 3.0 hit No. 1 on the AA video leaderboard
Two new image models now edit one spot and leave the rest alone
6/ Quick hits
HeyGen Video: built on MiniMax H3, from $0.01 per second at 480p through October
Higgsfield Genjutsu Restyle: restyle a whole video, keep the motion and camera
Comfy Agent (beta): builds ComfyUI workflows from plain language
Tripo P2.0: now officially launched