try using sdxl models
try using sdxl models
Vi am making consistent characters
Vits not about the models...this issue is there in all tthe models i tried be it sdxl or sd 1.5...
Vsee images and tell me ..what do you think
Vsorry for being rude.earlier
Sidk dude, this images looks realistic enough for me
Sdo you have any examples for what do you want to achieve?
Vdoes it not look plasticy to you..little bit plastic and the texture... all my friends said that they can tell its ai generated...
Vi was thinking of creating a realistic consistent character and try different clothes on them... i showed these images to my friends they said its okayish... they can tell that image looks ai generated.. that the texture on the image gives it away...
TDon't worry, can you give an example of the prompt you use? It may be that the problem is in the prompt, in the cfg (use a low one for realism), it is difficult to get a good skin texture, there are also loras that can help you
VPhoto of a handsome 30-year-old man with dark hair, stylish clothing (fitted blazer, black t-shirt, tailored jeans), standing near a stylish store on a bustling urban street at dusk, with soft bokeh in the background, shot on a Canon DSLR with a wide-angle lens <lora:ip-adapter-faceid-plusv2_sd15_lora:1> <lora:more_details:1> <lora:add_detail:1>
Vepic realism model + cfg less than 5 + number of steps 20
Vnegative prompt - (deformed iris, deformed pupils, semi-realistic, cgi, 3d, render, sketch, cartoon, drawing, anime, mutated hands and fingers:1.4), (deformed, distorted, disfigured:1.3), poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, disconnected limbs, mutation, mutated, ugly, disgusting, amputation
cartoon, painting, illustration, (worst quality, low quality, normal quality:2)
cartoon, painting, illustration, (worst quality, low quality, normal quality:2)
ADid they post this or did you Dr? Where can I read the rest of the thread? lol nvm found it lol
TI don't think it's a bad prompt, I thought maybe you were using terms like photorealistic and so on. Try this LoRA https://civitai.com/models/262452?modelVersionId=296756, it has given me good results but using it at lower levels. From cfg for realism and especially in epicrealism I think it used 3.5.
Tfor hires fix 4x_NMKD-Superscale-SP_178000_G
Vm using that also ..with 2x res and 0.33 strength'
KOkay after further testing you should 100% caption your images with supir
KWith Captions via kosmos
Kno captions just straight
TI also don't know very well how add Detail lora behave because I don't use them haha For realism I recommend leosamhelloworld5 or epicrealismXL (the last version merged with leo5), they are the ones that have given me the best results
Vthanks bro will check these models
Swhat is this kosmos thing?
SI just saw it
Sis there a local install for kosmos?
KYeah it’s the fastest captioning method
KDr furkan has a one click installer
KAnd it’s only 6gb too
Sdo you have a link?
Pim trying to instal Supir from installer but my speed is 749kiB/s] so all day pased and it didnt download everything. And next day it just started redownloading all files over....is ther ea way to download only files that didnt download?
Snvm, found it
L
L
Fproxy not working yet.
Fmodify the gradio app
Fedit output_dir = os.path.join("outputs")
Fset any place you want
RGotcha 
ROk 
RWho are we waiting on for graduó app?
Fcaptions better
K100% yeah.
Fyou know that is a local change. i can add later args parse
Fok let me try to add now
ROk!
RThen git pull
DHi, I installed v10 yesterday, and today I see new update v12 for Supir. Is there any way to upgrade it without having to delete and reinstall everything. You mentioned something about Git pull. Can you kindly explain what exactly I have to do to upgrade?
Fye
Fwait me
Fjust do git pull