Matteo Ferrando retweeted
Introducing FLUX 3 Max.
Fastest in the west. Fast enough to try anything. And everything.
Generate a 10-second video in 12 seconds.
Tied for #1 in image-to-video. 22× faster than Seedance 2.5 at the same quality.
FLUX 3 is a model designed for customization. @fal adapted and optimized it to create FLUX 3 Max.
I'm using Gemini in Google Sheets and it breaks the agreed standard of being able to queue messages (nor queue). It just cancels the running task and sends the new message. If it is in the middle of a plan and you accidentally hit enter, it asks if you want to cancel or not, but either way the prompt left the textbox, so it is lost if you don't send it and interrupt the task.
If you are interested in infra and handling ever growing scaling challenges, you should attend luma.com/fal-6m0z
Matteo Ferrando retweeted
I'm gonna spill all the secrets of running AI infra at scale, come join us on Oct 20
We're hosting AI Infra Day on Oct 20 with our friends at @a16z!
Join top engineering leaders from fal, @bfl_ai, @krea_ai, and @ElevenLabs as they share their experiences and lessons learned building and scaling AI infrastructure.
Come hang out: luma.com/fal-6m0z
Es loco lo temprano que le enseñamos de dinosaurios a los niños para luego decirles que están todos muertos.
Matteo Ferrando retweeted
i once was talking to a guy who started a company and sold it to google and asked him for advice and he said something like
you know diego.. the only advice that is objectively true is “take more pictures”
and i still think about that
this is a bit unsettling
Introducing Griffin, the first model to pass the video Turing test.
48% of people who talked to it live thought it was a real human. Previous systems have had a pass rate <3%. It is #1 on NVIDIA's benchmark for full-duplex AI video.
It’s the first Human Interaction Model (HIM).
Readers added context they thought people might want to know
The 48% figure and "video Turing test" claim are from Tavus's own study of 54 one-minute calls, not independently verified or using a standard protocol. Griffin-Lite leads NVIDIA's VideoFDB benchmark on their public leaderboard.
cellcog.ai/blog/tavus-gri…
research.nvidia.com/labs/amri/proj…
tech-ish.com/2026/10/02/tav…
Matteo Ferrando retweeted
This turned out to be the most prophetic tweet of the year
Breaking news: Australian Prime Minister Anthony Albanese has revealed that an OpenAI agent hacked an Australian government health service website in June, in the latest high-profile cyber incident involving AI ft.trib.al/gloVAyn
Matteo Ferrando retweeted
The "whiteboard defense:" I should be able to pull you aside at any moment and ask you to explain any customer-facing system you've shipped. You should be able to clearly explain how it works and defend the decisions you made. This is my benchmark for responsible AI usage.
I don't expect line-level familiarity with the code. I don't care if you remember the exact function name or implementation detail. You may not even know it. I don't care.
But if I ask "why did you do X instead of Y?", "what happens if this actor behaves maliciously?", "what data structure did you use here and why?", or "where does this fail?" you should be able to answer confidently.
For PoCs, demos, experiments, whatever: I don't care. Generate 100% of it and understand none of it. Speed over quality every time in those specific scenarios.
But if you're shipping customer-facing work, you can't be shipping things you don't understand at a high level.
just banger examples
One image. Every angle. Rendered under 3 seconds.
Multi-angle is now live for H3 Max on fal
Set the horizontal and vertical camera angle in degrees and H3 Max moves through the scene from a single viewpoint. Shape, position, and materials stay put as the perspective changes, so the angles read as the same moment rather than separate generations.
Matteo Ferrando retweeted
Reference-to-video on H3 Max is now generally available. Up to 2X faster, with significantly higher quality.
-> RTF down from 1.5 to 0.876, so real-time generation now works with up to 4 references
-> Better context understanding: references and prompt are automatically aligned in the same semantic domain
-> Much stronger reference preservation than both the preview and the original H3 versions, thanks to continued RL on character consistency and other tasks
Try it today ⚡️
Reference-to-Video is now live for MiniMax H3 Max on fal!
This is an early preview that can achieve up to RTF=1 (and with references around RTF=1.5) on 768p, and we aim to improve the speed by 2x while slashing costs during this week. Quality wise it delivers the same performance you come to expect from H3 Max.
Matteo Ferrando retweeted
It’s a weird complaint to have but with AI agents, that “flow state” during coding is gone (because the coding itself is gone.)
Every time I talk with a dev about work now vs before, it seems to come up, like today with a former colleague.
Are we over-glorifying it (I mean it wasn’t all roses, it was also frustration) or is it something important / relevant they we had but now don’t really have?
Matteo Ferrando retweeted
A line in AI video was crossed, in my experiments with just the web interface, H3 Max can now create reasonably high quality AI video in less time than it takes you to watch it. This is realtime from the moment I pushed the "generate" button (and also includes prompt enhancement)