Description
FAQ
Comments (9)
In more than 300 images with the exact prompt and settings used to generate the first image here, I've gotten some cute bear girls but they are nowhere close to having even a passing resemblance to Ambrosia. I got just as good results removing the lora and Ambrosia keywords entirely. Very dissapointed =/
what check point are you using?
@Warvilton I tried with Autism and Pony, they both give kind of hazy results. It starts off looking good, then it's like a blur filter is applied and things get a bit crooked by the end. I've found it helps to reduce the weight of the lora to around 0.5 to 0.6. Then add artist tags, like by bulletsoup, by shigoto, by hardiman. That seems to give better results.
Would love to try this for SD 1.5 if possible. Due to not having hardware to use Pony XL, and the sort of crazy x10 file size of pony Loras, compared to tiny 1.5 loras. Looks cute! Ty!
are you sure you're using a pony-based checkpoint ? I dunno about SDXL, but any pony-like seems to work
For the record: using autismmix confetti-mix, it's been very easy to generate ambrosia pictures.
I dont believe tags like 'by bulletsoup' and 'ambrosia' are even needed, but 'young female anthro bear, brown fur' sure might help get things into the right direction.
Is sad the fact that a midget got tagged as underaged
Works great with ponyxl, have made some submissions on a certain booru that have gone over well
ATF? or what Booru is it?