Watches always come unrealistic and faces are too smooth

https://ideasbugs.levels.io/photo-ai/p/download-photos-used-for-training-to-refine-training-set would help with allowing to change the training set with a new model at least

It used to be possible to download training photos to remove the wrong ones and retrain, but that does not seem to be possible anymore. Would it be possible to bring that back? That would make retraining a SD model on Flux easy as well.

My problem is each time I create a new model, they each have the same issue: the mouth is too wide and therefore the smile never looks like the original photos. I can keep retraining the model but without some way to adjust the too-wide smile, how will the new one be any better?

I love this idea as well as your response on Oct. 8. All good suggestions and questions!

Also really want to see this. For some reason my model ended up with wrong eye color and it's super annoying to have to re-find all the photos I used to train it all over again, because otherwise the model is working good.

Thank you. Does the model keep 'refining' itself as I use it? or it keep going back to use only the images that I originally used to train it? Does the images i 'Save' or put in 'Trash' have any effect of the model training? It seems like the model worked better before and now it gives a different person's face. It is also important to be able to view the images we trained the model with and be able to 'remove' and 'add' images before we re-train. Is there any benefit at all for re-train rather than start with a completely new model? Does the images for training all need to be in a certain size? Documentation or a short YouTube video will be highly appreciated. Thank you