Krea 2 LoRA training fits in 16GB VRAM on RTX 5080, user test finds

User test on a 16GB RTX 5080 confirms 768px Krea 2 LoRA training fits in VRAM, contradicting a widely-linked issue thread. 1,152 steps took 67 minutes at 3.42 s/it, peaking at 15,284 MiB (93.8%); turbo inference runs ~13 s per 768×1024 image.