I'm trying to use Gemini with a circuit projected on the BLS12-381 curve. My assumption is that the Gemini, being an "FFT-free" and more curve-agnostic proving system, I would be able to carry this out using application-specific curves like YAFA-108/146.

However, I'm having trouble fitting more than one instance when generating relations. This limitation happens during inputs padding here, which subsequently is caused by the field's low 2-adicity here.

Can the padding mechanism be adjusted to allow proving circuits projected on low two-adicity curves?