Is there a way to 100 model it accuretely based on the image’s mouth? Like how do I grab the vertices and accuretely match the image?
You don’t. An intelligent approximation is what you should model, as polygonal modeling is approximation of curves and surfaces.
If you’re asking how to refine the curve, use more vertices, possibly with subdivision surfaces.