2. Run with those settings and enable random warp and disable pretrain until SRC loss plateaus, on batch 16 and batch 8 for 8gb vram gpus until src loss plateaus
3. Turn on LRD and run until expressions in the preview are consistent, and DFM expressions are consistent, then switch the DST faceset to and then go back and forth between the DST facesets until it looks good with expressions, and you can check with a test DFM at this point.
4. Turn off RW, turn on blur out mask,keep LRD on. No hard and fast rules here, until it looks good as a dfm and doesn't flicker
5. Don't start Until all teeth and stuff are of good quality Start GAN with 0.1 power, 28 patch sizes, and 16 dims. Save and exit as soon as this starts; replace GAN files with the ones saved separately above.
6. Run GAN again with the same settings but watch closely - sharp face sets can be done in 15-30 minutes or a lot longer but varies and can be completely unusable with artifacts by 2 hours but varies
For the next SRC delete inter_AB, KEEP inter_B, and go back to the settings in step 2. Use the same GAN files originally put aside in step 1 each time when it comes back to step 5, reusing those further results in worse artifacts