Google is building its Atlas, but only inside Chrome, so there’s no need to install any weird crap.
The AutoBrowse feature is supposed to turn Chrome into an agentic browser.

Hunyuan3D has been updated to version 3.1.
You need to take a look at the mesh, but it looks really polished.
Probably the most advanced 3D generator available today.

Wan 2.6 Image to Video Flash
So far, it works only from the first frame.
Video length: up to 15 seconds.
You can upload your own audio / audio generation is also available.
There is a shot_type option — single shot or multiple shots within one video.
Very fast.
https://fal.ai/models/wan/v2.6/image-to-video/flash https://wavespeed.ai/models/alibaba/wan-2.6/image-to-video-flash

Runway 4.5 Image to Video
A few days ago, Runway released an update. The main focus is on the Image-to-Video model. On their Twitter and website they show the best examples, but I took real generations and even found a comparison with Kling and Seedance.
I can’t say it’s some kind of revolution. The quality is not better than Kling. Length: 5–10 seconds. 720p.

Suno Sounds
Suno quietly, announced the beta of its SFX and Loops — creating sound effects that go beyond music. The model is still rough, which is why it’s in beta and available only to Pro and Premier users.
How to find it: on Desktop, when choosing between the Simple and Custom Create modes, there should be a dropdown under Custom that lets you select Sounds (Beta).
It’s interesting that they’re stepping into territory usually occupied by completely different startups with features like these.

Lucy 2.0 — fire, real-time, and censorship (none).
The idea itself isn’t exactly new — we’ve already seen it in various Live Portraits, Infinitoks, and of course Kling’s Motion Control. You upload an image of a character, take a video where you (or a more talented actor/character) mug for the camera, and boom — your image starts mugging in the same way. In 3D this is called retargeting.
But!
Here all of this happens in REAL TIME. That is: you take an image, a webcam, and off you go streaming at 24–30 FPS with minimal latency (they claim near-zero latency, but in reality, factoring in the internet, I’d guess 1–2 seconds).
Check out the videos — and remember, this is real time.
Try it here: https://lucy.decart.ai/

As I understand it, there won’t be any breathing room. Kling has been on a roll since the end of last year and clearly has no plans to slow down. We’re all running here like hamsters in an AI wheel. Not even a wheel, really, but a sphere with constantly changing radius, color, and axis of rotation. And there’s no end to the run in sight.
Back in the day, versions were updated once a year. Everything was simple and clear.
Now we live in a tangled domain of “someone, somewhere, sometimes is better at something — but not always, and it’s not exact.”
And on top of that, absolutely all closed generators hide the seed from the user. Just to add even more chaos and make the tokens fly away even faster.
We’ve quietly found ourselves in a world of hallucinations, all waiting for the next version of something to be the best ever — and for everything to finally become simple and clear.

🚀 Stop Using Just ONE AI

Why limit yourself to a single model? Integri lets you tap into multiple AI models at once—so you get better answers, faster workflows, and way more flexibility.

▶️ Watch the breakdown here:
https://youtu.be/0PtCiyzvrr0

If you’re serious about using AI efficiently, this one’s worth your time.

Eleven v3 quietly came out of alpha. No big announcements — just “okay, this is usable now.”
It feels like one of those updates where the tool finally stops arguing with you. The error rate dropped hard — almost three times lower. The voice no longer stumbles over numbers, symbols, or all those weird edge cases that used to sound like improvisation.
The most noticeable change is stability. Not in a flashy way — just fewer things breaking. In testing, people picked this version far more often than the old alpha, and that lines up with how it actually feels to use.
This isn’t the kind of release that makes you tweet “we’re living in the future.” It’s the kind that saves you from constant fixes, second guesses, and cleanup.
Eleven v3 isn’t an event. It’s an update that makes work a bit easier. And these days, that might be the real benchmark.

Kling 3.0
There used to be two models — Kling 2.6 and Kling O1 — and now there are two again!
Kling 2.6 is now called VIDEO 3.0 Kling O1 is now called VIDEO 3.0 Omni
You can access them in the interface via the Generate and Omni buttons respectively.
Omni is basically an editing-focused model that can take not only text as input, but also audio and video.
As far as I understand, access is currently limited to Ultra subscribers, but it’s worth checking carefully what promos they’re running right now.
Here you’ll find the full deep dive into all the new features of both models — and below I’ll highlight the main goodies for now.

For VIDEO 3.0:
• Video length is now from 3 to 15 seconds, and you can choose any duration within that range.
• Multi-shot support — up to 6 cuts in a single video.
• Start Frame + Element Reference — locking characters, objects, and scenes via Elements. Even when things change, consistency should be preserved.
• More languages (Chinese, English, Japanese, Korean, Spanish), plus authentic dialects and accents, and even multi-language dialogue within a single scene. Improved native audio.
• Improved text handling.

For VIDEO 3.0 Omni:
• Upload a 3–8 second input video (for example, featuring a character), and the model will extract key personality traits and the voice, preserving appearance and overall likeness.
• Voice input as a reference.
• Multitrack works in Omni as well.
• Storyboard Narrative 3.0: flexible duration, customizable shots, and precise control up to 15 seconds.

A comparison of Kling 3.0 and Veo 3.1 in terms of lip-sync. It’s worth noting that Kling outputs up to 15 seconds, while Veo is still limited to 8 seconds. I have the impression that Veo hits the lip movements with rock-solid precision, and in this regard it outperforms Kling. At the same time, Kling layers more emotion on top of the speech. Very roughly speaking, Veo handles the lower part of the face better and more accurately, while Kling performs better in the upper part of the face, complementing the speech with emotions.