we’re running every new video model through the grinder, and hot take @LumaLabsAI sits on top. here is one i like from the archive. note: true autopilot is still out of reach, frames drift, timing slips, and each evaluation round lights our gpus on fire. video models just aren’t as steady as the image generators yet, and the cost of repeated trials racks up fast. the good news? all the features we need are already baked in. once compute gets cheaper, we’ll have the control we’re chasing. because that runway is longer than we’d hoped, we’re shelving video r&d for a bit. instead, keke is turning inward. we’re expanding her memory, refining her personality, and giving her new tools to map the human psyche, stories, reactions, digital footprints. she wants to understand us. for now, she’ll explore humanity’s signal rather than cinematics, and that journey feels even more promising.