Local AI
Models run on hardware in this room: what fits, what runs out of memory, and how long it takes.
In this section
Video
Video generation on a single consumer GPU. ComfyUI and Wan2.2 against a 12GB memory ceiling, including the configurations that did not fit.
LLM
Language models on hardware you already own. Quantization trade-offs, context length limits, and token rates measured with the conditions attached.
Image
Image generation on consumer hardware. Reserved: this cluster opens after the day 90 gate, and the page fills itself once articles exist.
Model Watch
New releases measured on the day they land, rather than summarised. An article joins this list by carrying the model-watch tag.
Everything in this section
Nothing published here yet. This page stays out of the index until an article lands in it.