What models are necessary to obtain face embeddings?

First of all, thank you for this great library. I'm attempting to speed up the calculation of embeddings for faces, but I'm somewhat confused. If I use allowed_modules=['recognition', 'detection'], it indeed results in an approximate 30% speedup, and the vectors seem to be the same (within a range of approximately 1e-5). However, I had assumed that the algorithm for embeddings used alignment, relying on a model that detects key points.
So, it appears that only two models are necessary to obtain embeddings: recognition and detection. Is it correct to understand that no alignment is performed by default?
The code I'm using:

```
from insightface.app import FaceAnalysis

allowed_modules = ['recognition', 'detection']
app = FaceAnalysis(name='antelopev2', providers=['CUDAExecutionProvider'],
allowed_modules=allowed_modules)
app.prepare(ctx_id=1)

t = time()
for _ in tqdm(range(1000)):
res = app.get(img_array)

print(time()-t)
```