(1.07 MB 1024x1024
05795-288590622-1girl,(gyaru_1.3),dark skin,, school uniform,, cleavage, small_breasts, miniskirt, smile, solo, ,,____cardigan,__plump,.png)
>>11715
In those images, I used a combination of the "fat" lora and the drworm one, with the Anything-v4.5 model. The trick I've learned for text to image using the drworm lora is to use the PLMS sampler, and *make sure* to upscale by 2 to avoid having garbled faces, with 30 steps to get rid of all the black splotches (rarely, you just need to put in a different seed, since the splotches remain). DivineElegance also works as a base model.
But the way I usually make images like picrel now is to generate a base image *without* the drworm lora, using e.g. DivineElegance (any sampler, though I prefer DMP ++2M Karras), then put that image in img2img, add the drworm lora with a strength of 1 (optionally removing other artist loras in the prompt), switch the model to ChubbyGirlsGolden, then run img2img with the Euler a sampler to convert it into a drworm style.
I might actually try making a guide at some point, but I hope this explanation helps.