Clear guides, hands-on experiments, and useful tools for people who create with AI and build their own setups.
Different goals. One practical point of view. Start with what you want to make, run, or understand.
Prompts, visual workflows, and practical methods for turning an idea into a finished creative result.
Local models, consumer GPUs, workstation planning, cooling, power, and hands-on troubleshooting.
Useful AI tools, practical comparisons, reusable resources, and workflows worth keeping.
A 5-second 864×480 MiniMax H3 clip took 19 min 29 s on a Tesla V100-SXM2-32GB. A 16GB V100 finished in 20 min 15 s. An RTX 5060 Ti 16GB did the same size in about 294 s.
MiniMax H3's ComfyUI model files total 39.55 GB. On a modified 32GB RTX 4080 SUPER, 864×480 clips took about 148 s and a 1920×1088 clip took 53 minutes without running out of memory.
MiniMax H3's model files total 39.55 GB. On an RTX 5060 Ti 16GB with 32 GB of RAM, 864×480 clips took about 294 s, and a 5-second 1920×1088 clip finished in about 72 minutes.
We moved an RTX 5060 Ti 16GB from an old X99 workstation to a Z790 DDR5 PC. A 27B model that spills out of VRAM sped up 164%; a 12B model that fits gained 18%.
AI moves fast. Good decisions still come from understanding the task, testing the tools, and knowing the trade-offs.
What we're building