LoRA alone caps near 85 percent identity match. Stack IPAdapter FaceID v2 on top and you cross 95. The exact node graph and the failures to avoid.
The IPAdapter weight that holds identity at portrait range breaks at full body. Here is the 100-image grid that shows the safe range per shot type.
Turn one selfie or render into a full turnaround sheet the AI can lock to. Front, three-quarter, side, back, plus expression strip. Real workflow.
A real workflow for getting the same face, same outfit, same vibe across an entire image set. Without character drift, without re-prompting fifty times