Hi lewnus,
Your object/face has probably moved slightly between the different images, hence PhotoScan is struggling to find good quality tie points (PhotoScan's SfM algorithms assume that the object is stationary/doesn't change). If you want to get best possible results for face reconstructions, you would need to use a multi-camera setup with synchronized cameras.
Anyway, here are some things you could try though:
1. Use HIGH accuracy for photo alignment which guarantees that pictures are used at full resolution (only works if the quality of the pictures is good). If the quality of the pictures is not great, you might get better results using LOW or MEDIUM.
2. Manually add markers to 'force' PhotoScan to align the images
3. Take new images and tell your model not to move/breathe when you take the pictures.
4. Increase the distance to the object/face. Leads to lower resolution, but probably better image alignment results in your case.
If these suggestions don't improve your results, then please share the project and I'll have a look at it.
Regards
SAV