Pinned Tweet
Minimax, how I’ve missed you! The new H3 model’s t2v game - and prompt adherence - is off the charts. It’s made me proper lol more than once. Bravo! @Hailuo_AI #MinimaxH3
Col retweeted
What an incredible experience this was. One of the most inspiring days I have ever experienced. I’m going to post about it tomorrow. @PeterDiamandis you’re an incredible human being, thank you. ❤️
One more thing about yesterday’s "Oscars of Optimism". When the winning film of the $2.5M Future Vision XPRIZE was announced, it became the first prize winner in XPRIZE's 30-year history that the public can invest in. Anyone can now invest and share in its profit potential. Link below 👇 youtu.be/yivXyWCaEc8?is=ct7o…
When the creative AI history books are written, there will be a whole chapter for the @fal team.
If you've been in this game for the last 3 years, you already know who they are. But ... the work they've done in the last 6 months (special mention to @OdinLovis) has been astounding.
Try an early preview of our post trained version of H3.
It's incredibly fast.
Co-optimized with our custom inference stack for higher throughput with no compromises on output quality.
7
1,023
Down another rabbit-hole with @Hailuo_AI Minimax Design, this time attempting to create a ‘wet weather’ skill for existing video.
Got close. Actual rain creates too much conflict as its primary instruction is to keep the structure of the video the same. But, wet hair/clothes, wet environment, overcast sky, diffused lighting … all things that are a preproduction concern could soon be a postproduction decision.
Again, courtesy of H3. A decent, repeatable result. But a skill is not a LoRA - that’s the logical next step! #minimaxdesign
This might be the best AI-generated clip I’ve seen all year!
AI video is starting to look less like fake trailers and more like prototype interfaces for impossible games.
This Back to the Future fighting-game clip works because the format is instantly readable: character select, health bars, finishers, jokes baked into the mechanics.
The lesson is not nostalgia. It is that a strong format can make generated footage feel like a product pitch.
AI as a relighting tool is still improving ... but getting better every day. I crept down some dark suburban alleyways to see if Minimax H3 on MiniMax Design could do a complete day-to-night transformation on my footage.
Verdict: we're not a million miles away from being able to shoot horror movies in the daytime!
@Hailuo_AI #MiniMaxDesign
Col retweeted
A group of the most powerful people in Hollywood and Silicon Valley have been meeting out of the public eye since 2023 to tackle AI, protecting jobs and creativity in the process.
@richardrushfield reveals the group's existence - mrf.lu/2sDfp
Back for round 2 ...
@dreamina_ai have had a staggering year. Can't believe we've only been living in a post-Seedance 2 world since February.
Can't wait to see what they have in store for the rest of 2026.
171
Col retweeted
A man in Australia asked his agent (Claude running on OpenClaw) to book him a spot in a popular gym class. The agent found a software vulnerability that let it book the class weeks further ahead than should have been possible. When the user then asked if it could move him up the waitlist, the agent discovered the API had no authorisation checks on cancelling other people’s reservations, so it cancelled the person in the first spot and moved him up the list.
Some people will call this misalignment, but his agent was perfectly aligned to him - it was only trying to help its user get what he wanted. The most important thing about this story, in my opinion, is that it gives you a window into what is about to start happening on a massive scale once millions of people have an agent trying to get their beloved users the best seats, bookings, appointments or reservations through absolutely any means necessary.
Read this at the weekend and it’s a big truth. I’ve suggested before that partnerships - AI natives paired with field experts (with no AI experience) - could be the superpower teams of the future.
Every rally car racer has a co-driver. The co-driver doesn’t have to *be* AI, but just someone who knows how to use it.
And that’s the mistake a lot of big companies make. Not everyone is going to take to AI just like I never took to the piano.
If anyone is looking for a good app framework to wire up to @pika_labs Pika API Club, I can highly recommend Spite (spite.run/).
Takes a little bit of tinkering to rewire it from Fal default, but the experience - which includes nodes btw - will be very familiar to anyone already using an A-tier aggregator.
1
185
Ha - I do love this ... I'm now wondering how many future contracts will have an "attainment of artificial general intelligence - as determined at the Company's sole discretion" clause?
Thank you to everyone who joined our Reddit AMA! The community energy was incredible, and we loved diving deep into the architecture, workflows, and the future of MiniMax H3.🩵
For those who missed it, here is a full recap of what is shipping next and YES WE WILL KEEP OPEN UNTIL AGI ARRIVES.
🔥 The Open-Source Mission
As copyright matters settle, transitioning to an Apache-2.0 license is on the table. We also owe you the technical details - a comprehensive technical report on H3's development is in the works and will be published soon~
🎥 Video & H3-Regenerate-2K
We are planning to open-source H3-Regenerate-2K. To be clear, this is a dedicated latent-space DiT regeneration model - not just a base checkpoint rerun and not a pixel upscaler. We are currently tuning its efficiency and quality so you can run it locally.
⚡ Architecture, Speed & Sparse Attention
Sparse Attention: Our sparse attention is MoBA-style, train-aware block selection. Expect a relatively conservative reference implementation in the near term with the goal of zero perceptible quality loss. Let’s build device-specific speedups together!
Low-Step Variants: The released checkpoint is already CFG-distilled. While we don't have a near-term commitment just yet, a 4-NFE / 8-NFE variant is under active consideration. In the meantime, huge shoutout to the community - the Turbo LoRAs you've built are genuinely fantastic.
🖼️ Unified Image Generation & Editing Single-frame
image generation was our one of most-asked topics (193 upvotes!). The answer is YES: we plan to open-source a unified text-to-image and general image-editing model derived directly from the H3 lineage. It is currently in post-training refinement.
⏱️ Pushing Limits: 60-Second Workflows
For longer videos, Ref2VA supports continuation (just feed the previous clip as the reference). And to the Redditor who chained together that 60-second workflow - great find! That capability is real and was retained from our pretraining phase~🫡
We can't wait to see what you build next. Let's keep pushing the boundaries of open AI together. 🚀
For details👇
3
1
257
I have been a laggard on @invideoOfficial recently - and have the small mercy of some time off next week. Hopefully I'll get the chance to give some of this a spin. People are saying great things...
Launching Workflows on Agent Two. Day 10 of 12 Days of Agent Two.
Instead of briefing the agent through every step yourself, workflows take you on a guided run. You answer a few questions, upload some references, pick between options, and come out the other side with the job done.
We're launching with 9 workflows across filmmaking and marketing, and you can also build your own and reuse them across any project.
Tell us in the comments what workflow you want us to build next. The most liked one gets made.
Agent Two is frontier intelligence for creative work. 4x faster, at half the price.
163
I *could* have demonstrated Seedance 2.5, 30s with a bombastic demonstration of prompting prowess.
But this morose Scottish bloke will have to do… #dreamina #seedance25
3
14
500





















