Romantic light and color illustration
Version 1.3 notes: this model started by training on some classic illustrators (JC Leyendecker and Maxfield Parrish prominently) but quickly moved more digital, still taking something of the DNA of classic illustration, especially in terms of romantic light and coloring. Initial versions suffer from some awful drawing at times.
this latest version marginally improves drawing of hands and faces (roughly about as consistent as base SDXL without LoRAs or Refiner) but greatly improves on consistency of lighting and romanticism.
For the most part, you needn't even use any keyword. But if your imagery turns too photographic, insert 'illustration' somewhere and it should get back on track.
I don't know if I can do more to improve on the hands ... but if I can there will be future versions of this LoRA still to come.
Description
We're done with beta versions - this is 1-point-OH!
FAQ
Comments (9)
Amazing style.
thanks so much! It may see another version update if I can polish up some of its wonkier elements. But I won't upload anything if it's at the expense of the current overall aesthetic.
which sdxl model was this trained on? just base?
the base model, yes. That's always my approach, to maximize compatibility. But I also have a strong personal preference for the base model, using LoRAs to reach specific styles rather than custom checkpoints. Any custom checkpoint, even as it refines some aspect, is destructive of the initial SDXL's versatility. The more valuable custom checkpoints, in my mind, are anime - but that's just wholly outside my personal interests. So for me, it's all about the base model!
This is great, do you have SD 1.5 version available?
sorry - I've somehow just never had any luck training for SD1.5 so far. I'll keep giving it a go, but there must be a crucial gap in my understanding of LoRA training. Everything I've tried has turned out to be pure garbage.
@eldritchadam Can you tell me which large model you used to train this LoRA? I would like to make an attempt, thank you
@ccccjmg777 Base SDXL. I'm something of a purist about it. I hardly ever use the fine tuned models. I don't need either the anime or the NSFW content that most fine tunes are largely trying to improve on.
@eldritchadam I understand, thank you for your reply