On 12 October we could make about one minute of consistent AI film. Today it is 12 minutes. Same mule in every shot. One narrator. Real-time motion. Made locally, on our own GPUs, with open models. Next: 30 minutes. Then three 30-minute segments for 90, and on to 120.

Measured on MiniMax H3 across two DGX Sparks: two boxes don't make one clip faster. Each renders its own share at the same moment, so you get two clips in the time of one. Treat a second GPU box as throughput, not speed.

Mistake of the week: one Mac had two Tailscale daemons and I stopped the wrong one. Two minutes off the network, two mounts dropped. Every change now waits for the system to settle before checking itself, and rolls back if the check fails.

We just published the plan: @TheModelMule backers book scheduled time on our own NVIDIA GB300 instead of burning tokens. Our measured cost: $3.10 per finished minute of video, against $6 to $35 on a meter. Applications are open now. xitter.cf/TheModelMule/status/21… #ModelMule #GB300

Our two GB10s kept hard powering off mid-render with nothing in the logs. Cause: clock spikes tripping the PSU overcurrent protection. Fix: cap clocks with nvidia-smi -lgc (power limit isn't supported on GB10). Capped, MiniMax H3 ran about 8% faster, not slower.

Hey @X! I’m at 843 followers, aiming high to hit 3k 🚀 I’m building ModelMule, a non-profit IRC hub for open-weight local AI models, with a simple interface for people new to IRC. Keep Local AI Alive 🔥 Next up: bringing a GB300 online for compute-constrained folks doing serious AI development. Looking for 600 backers at $1k each to launch with unmetered access ⚡ Help me #connect 🤝 with people who get it: 🦙 Uncensored weights > API keys 🔥 Your GPU is too small and you know it 📡 IRC never died, it was waiting for us 🐧 Omarchy or nothing 🛠️ Building real infra in public Follower count doesn’t matter, good conversations do! ✨ ko-fi.com/modelmule