Staying In, Red.
NSFWBonus Images.
parameters
masterpiece, detailed, beautiful, oil painting, 1girl, laying down, from above, mature female, milf, short blonde hair, laced hair, gold eyes, gorgeous tight red dress, pearl necklace, seductive, revealing, cleavage, large breasts, golden tiara, simple background. adult, age 30.
[ Photo, ((high detail skin)), <lora:GoodHands-vanilla:0.5>, <lora:epiCRealismHelper:0.5>.. ((best quality)), ((masterpiece)), (detailed), ultra quality, highly detailed, realistic, high quality, 4k, 8k, Meticulous design.]
Negative prompt: (worst quality, low quality, bad quality:1.3), embedding:ng_deepnegative_v1_75t, embedding:EasyNegative, embedding:badhandv4, minor, child, immature.
Steps: 30, Sampler: DPM++ 3M SDE Exponential, CFG scale: 5.5, Seed: 737986088, Size: 512x700, Model hash: a594874016, Model: realcartoonRealistic_v12, Lora hashes: "GoodHands-vanilla: c05ed279295e, epiCRealismHelper: c5e24d9bc4b5", TI hashes: "ng_deepnegative_v1_75t: 54e7e4826d53", Version: v1.7.0
postprocessing
Postprocess upscale by: 4, Postprocess upscaler: ESRGAN_4x
note: original prompt taken from this civitai post and modified
It’s locally run stable diffusion 1.5 with a few models and loras, nothing too special. Using an old 1080ti it’s more than enough.
Most of the work here is being done by the model
How do you make the faces turn out so well? Given you only posted 4 pictures but the faces are flawless on all of them. I’d say my yields are lower than 25% but it would still be hard and take a while to come up with 4 pictures of that quality. More then 50% of them turn out pretty good but on a lot of them the eyes or face is always messed up just enough that no amount of face fixer passes makes them acceptable.
I find the best chance of having a recognisable face is to have the subject facing the viewer. With subject looking away or at a weird angle it has a hard time for sure. Have part of your prompt reference eyes or face “gold eyes” or “portrait” may help.
Turning on face restore helps.
Embeddings and Loras also help to some extent.
embedding: ng_deepnegative_v1_75t / EasyNegative / badhandv4,
lora : epiCRealismHelper
Models: I am finding the “realcartoon” set to be pretty reliable.
You can also try adetailer, which after the base image is generated will re-generate in more detail either the face or body (your choice). Note: I am not using this at the moment as it slows down generation considerably.
What model(s) are you using? Does the same happen if you copy this prompt for example? Happy to try one of your prompts out to see what the result is.
Edit: final thought, some models struggle with resolutions they weren’t trained in, so stick to 512x512 unless you’re sure then upscale your choice.
I don’t have the same loras as you, but reusing all the parts of the prompt that are relevant gets me not bad pictures. Maybe I need to use more words in the prompts.
I use this one called sdxxxl_30 for making nudes and realismEngineSDXL_30VAE for everything else. realism engine generates an image closer to this when using the prompt.
I don’t subscribe to the 512x512 thing. I’m sure there really are technical issues with not doing 512x512 all the time but 768x768 gets better quality just because there are more pixels on the eyes which reduces the frequency of getting fucked up eyes. Doing 768x768 instead of 512x512 increases yields quite a bit and definitely offers more benefit than it does harm in my experience.
hmm, the “upload image” button is broken right now. It uploads but doesn’t give me the url where the image is so I can’t actually put it in my comment (or on a post, hope it gets fixed and isn’t a problem only I have because I have a porn album to upload at some point)