Diffusion & world model infrastructure. You create, we make it faster and cheaper with NO quality loss. Video, image & audio models via API, or self-served.
San Francisco
Joined March 2026
- Tweets93
- Following383
- Followers546
- Likes319
MachGen AI retweeted
MiniMax H3 was open-sourced two months ago and the @MachgenAI team has been busy building cool features on top of it.
H3 is no longer only one model but a collection of different pipelines & accelerated LoRAs. Sharing a few things that our engineering team built on top of Minimax-H3 🧵
MachGen AI retweeted
Okay, MachGen just made MiniMax H3 way more interesting.
H3 normally gives you 15 seconds, but I just pushed a single generation to a full 30 seconds on MachGen.
No stitching clips together. No awkward cut halfway through.
Just one continuous shot with enough time for the scene to actually develop into a story.
I tested it with this little story and the consistency genuinely impressed me 👇
30 seconds from one idea feels completely different.
Try it yourself at machgen.ai
MachGen AI retweeted
"Diffusion models aren't LLMs. So why run them on LLM infrastructure?"
-- Kismat Singh on @MachgenAI building a diffusion-native stack.
A few inference benchmarks with MachGen:-
→ Flux: 6.2s → 1.5s
→ Wan 2.2: 98s → 16.5s
→ LTX 2.3: 67s → 10.7s
Checkout the complete interview with @UniteAi - unite.ai/kismat-singh-co-fou…
MachGen AI retweeted
broskis @MachgenAI have free Minimax h3 turbo free gen if your account has $25+ balance
time to make some vids then
🤝 Paid partnership
