1 mb LORA trained in 5 mins that does the same thing as 2.5 gb model but better

I made this LORA to prove my point made here. This LORA does the same thing as this model. but better since it's much more portable, 1975 times smaller in size, more versatile and doesn’t require merging and you can easily adjust the strength on the effect.

The training was done on my phone, and literally took 5 minutes and 10 seconds, the prep for the training took another 5 minutes, i simply used the 6 examples provided in the model above for training, no captioning or anything else, i used Luisap method mentioned in my post above.

for the examples i included, some had no prompt engineering whatsoever like in the original model, some had engineering and used other LORAs.

Please stop making checkpoints for very specific concepts when there are other more suitable options unless there's a very good reason why you should. also when training LORAs for a single character, consIder adding other characters and maybe a style too, to make the LORA more useful.

This is not a direct criticism of the creator of the original model (he works really hard and produces some really cool results, check them, also this is for 1.5 not 2.1), or anyone else, but i think if more people did this, it could really improve the output on this website in my opinion with much more efficient models, that have smaller sizes and can do more.

Thanks.

notes: the strength really depends on how closely you want the images to look like the examples in the original, 0.5-0.7 is probably a good range. and i recommend lower steps (12-18) with highres fix

trigger word isn't necessary.

and just to be clear, this isn't an extraction.

Also admittedly the title is a bit clickbait-y since it's not really meant as a direct comparison to the original model really, it's just that I really feel more people need to see this.