NEW MODEL ADDED SEPT 22. 2023
(Green Text = Updated or new Information)
FerVency (V1) & (V2)
This Model comes in 4 Versions
1) FerVency - Photo Realistic Photography
2) FerVency Glam - Everything is a bit over the top and more alive
3) FerVency Cell Phone Pics - Pictures resemble images taken with your phone
4) FerVency (V2)(NEW) - Contains the best features of the 3 V1 Versions (faster Performance)
What's New in V2
FerVency V2 is like a good old family dog, it does exactly what you tell it to do and it gives you love all the time. Basically it contains the best features of all 3 previous versions but spits out way more realistic results and is performance improved. Especially when you are using a Highres Fix and a refiner and you step out of the room to render a batch of 20 images, you will find way more result sin your output folder as you have expected.
If you liked specific characteristics in one of the three V1 versions, still use FerVency (V2) and use your previous favorite as a refiner. You should like the result.
While the previous three versions were aiming or a particular set of images, you will be able to get way more out of V2 even High End Portraits by famous photographers
First of all, let me be clear that I deserve little to no credit, I just had a ton of fun merging things until they were finally consistent. There are amazing models out there for very talented people, what I wanted to for myself was to create consistent images and styles for AI Influencers so I can place them wherever I want, wear cool and unusual colorful outfits and present them as if they were taking selfies or in a high end shoot by a famous photographer. You'll like how consistent the faces and bodies will be and if you prompt some of the non ?????? shinier outfits like satin, spandex, lycra... you'll get a lot out of it.
HOW TO USE FERVENCY?
all 4 Merges are put together so you can use the exact same prompt and get a reliable result. If you want to crank out a huge batch and upscale your favorites afterwards, don't do it IMG to IMG, go for PNG INFO of your favorite and generate it off of the prompt.
Steps: 30 to 35
CFG: 3.5 to 7 depending on how much saturation, freckles etc. you want.
Best Size 512 x 768 or 768 x 1024 if you have a 3090 or higher
Best Samplers: DPM++ 2M SDE HEUN KARRAS or DPM++ 2M SDE KARRAS
In Version V2 you will have more freedom with the samplers and get similar results, feel free to choose the one with the best performance
Hires Fix:
UPDATE September 19th 2023
UPSCLAER: ESRGAN_4X (For Faster but good quality output)
(Resize Steps 15 - Denoise 0.20.- 024
Hires Checkpoint: Fervency V2
UPSCALER: 4x_NMKD-Superscale-SP178000_G.pth (Better Quality, takes longer)
(Resize Steps 7 to 10, Denoise 0.17-0.33)
Hires Checkpoint: Fervency V2
REFINER: If you are using one of the three V1 Versions, add FerVency V2 as a refiner, if you are using FerVency V2 and the result is too saturated use Epic Dream
Switch at 0.8 is what I use by default but there is no wrong or right here
USE VAE of your choice, V2 is pruned set to automatic
The best way to control these models the best is by simply describing what you want, you will end up getting a good basic result. From there weight things, you want smaller ???? and bigger ??????? just write (narrow ????:1.1) or (????:0.9) and so on. You this like a slider from 0.5 to 1.5 and if your prompt doesn't take make sure you have no conflicting information and if that still doesn work, make it a priority by using double brackets.
Loras
Most Loras will distort your results, you have to weigh them in, 0.5 to 0.7 seems to work best. Many... if not the most will mess with your models face because they were trained with a different face, sometimes you can fix it with clip skip and if that doesn't work just use roop or face swap and write your girls face over the trained face.
Weird Faces
Especially from side angles and long distances you will keep running into weird faces, Hires Fix will 9/10 times fix this, if that still doesn't work use roop or face swap in your chain.
New Girls / Changing Outfits
You will find that you get a good portion of new faces out of these models, that being said, SD loves to change them, slightly and evolves from that, The outfits is similar but different... use ControlNet /Reference (no model needed) for you next 20 generations, you will find more constant results. More consistency in clothes + more options in (V2)
NSFW Content / Loras / Embeddings
when limbs get stretched or your batch produces consistently too many limbs or completely produces inhuman results, reduce the size to 512x512
