try samtools and bcfools to pick consensus fastq then convert fastq to fasta using seqtk

samtools mpileup -uf REFERENCE.fasta SAMPLE.bam | bcftools call -c | vcfutils.pl vcf2fq > output_cns.fastq

seqtk seq -aQ64 -q20 -n N output_cns.fastq > output_cns.fasta

The above seqtk is to filter the bases of quality lower than 20 to N. You may adjust it according to aim of your study

You're likely not getting variants called due to not adjusting the 'max read length' parameter, since valid variant won't be called on read lengths >500 by default in bcftools.

That being said, you shouldn't use bcftools for doing long-read variant calling since the error rate profile is significantly different between long- and short-reads, the latter which bcftools has been validated and designed for. I haven't used long-reads for variant calling purposes quite yet, but Medaka has support for long-read variant calling; however, I would note that depending on what organism you are sequencing, you'd have to train their neural network classifier to ensure it's calling variants correctly. I would recommend using a control to test in that case whether real variants are being identified correctly.