Is it possible to implement these features from ai toolkit into kohya. Fully explained down.

I have done over 300+ trainings now (renting gpus) of different training parameters from both kohya and ai toolkit.

I get good to decent results now on both.

I believe if you implement these kohya can improve. Im not sure if im right or wrong, but please let me know.

1. noise_scheduler: flowmatch ||||||||| with the sampler - sampler: flowmatch
not sure what this does
2. content_or_style: balanced
seems to give a good balance when prompting
3. quantize: true
i tested quantize and the quality is basically close to the same, but with even lower vram usage to use.

4. multires training
hopefully an easier way to implement it where it just takes your one dataset and buckets it into 512,768,1024 or whichever res it needs to bucket correctly.

5. linear_timesteps: true
not sure what this does but apprently it improves the model

The issue names no files, tests, or entry points. Start by investigating how sd-scripts currently handles noise scheduling, samplers, quantization, content/style balance, and multiresolution training; completion would require clearly scoped, validated implementations for the requested options and evidence that training quality or VRAM usage improves.

Written by the indexing model from the issue text.