Big Update: Version 2.0:
I played around with a lot of Loras and checkpoints. The model has changed dramatically. I don't remeber what i have merged in ;-)... sorry for that.
As a result you get a model that can produce very nice results.
My favourite settings are cfg 1 and 8 or 10 steps. Using cfg 1,5 you get more realistic results. For Comic and Arts use cfg 3.
Summary of 1.1 and 1.2:
A combination of my favorite Flux Checkpoints and Loras. It is based on Flux1.S so you can get good results already with 4 steps. I uploaded Version 1.1 because I realized the problems of 1.0 early, so i didn't uploaded it.
("flott": german, translation: "fast")
Version 1.2
only a small update. I merged the Lora "NSFW_MASTER_FLUX_-_v1-0". The effect was massiv, as you can see in the comparisson examples. The checkpoint is much better in NSFW now. But a little - really a little - less good at art.
Version 1.1
Details:
Merged checkpoints:
Flux Fusion DS -Smooth merge- -4+ steps- -AIO - UNET- -ALL GGUF • NF4 • FP8-FP16- - v0-fp8-e4m3fn -AIO-
ArtFusion Flux 4-steps - v1-2
The Araminta Experiment -SDXL+Flux- - Flux1-A1
SPEED FP8 - e5m2
matrix_v10
Loras:
Full Female Anatomy - Fineart Nudity - Flux Lora -NSFW- -Breasts-Pussy-Ass-and more - v1-0
SCG-Anatomy-Flux1-d - NSFW LoRA for Flux- - scg-anatomy-female-v2
Black and color - v1-0
Flux Detailer - v1-0fgfg
Settings:
CFG: 1 --> min. 4 Steps
CFG: 1,5 --> min. 8 Steps
CFG: 3 --> min 30 Steps
Sampler: Euler Beta
Examples:
prompt: a modern logo with the word "SCHNELL" and a cozy photorealistic cat sits at the side
Description
V2 is a big step. I played around with a lot of Loras and checkpoints. The model has changed dramatically. I don't remeber what i have merged in ;-)... sorry for that.
As a result you get a model that can produce very nice results.
My favourite settings are cfg 1 and 8 or 10 steps. Using cfg 1,5 you get more realistic results. For Comic and Arts use cfg 3.
FAQ
Comments (13)
@Guntwin 8-10 steps and CFG up to 3? Is this truly 100% Schnell-based or is there anything from Flux Dev mixed in? Unfortunately. if there's even a tiny amount of Dev then you'll need to change the license to the non-commercial version.
In any case, I was impressed by v1.2 so I'm looking forward to trying this.
i don't understand the behavior of this model, but everything i megred was labeled Flux.S
but thanks for the tip. To be on the safe side, I have changed the version to Flux.D.
@Guntwin I just downloaded it and it definitely has enough Dev in it to require the full 20 steps to fully converge. You can get away with 10 but 20 is needed for best quality. In the next version it might be a good idea to also merge in the ByteDance Hyper Lora so that it will work best in only 8 steps.
That said, as it is it makes very nice images!
@Grumblebutt thanks for the hints
@Guntwin No problem, thanks for the model!
and again... i dont understand this model. I used the lora... pixel noise... i tried to merge it... and again... it doesn't work with this model. I think i need to rebuild my next model from the ground
@Guntwin I don't know much about creating models 🤷, but as long as the results look good then mission accomplished
@Grumblebutt the same i think about it. I have experimented a lot with comfyui... try it, it is simple, if you don't fear it ;-)
@Guntwin Thanks, I'm very familiar with Comfy but haven't tried making models yet. I do need to try it though.
I have a very technical question (ahem): how on earth do you get SMALL breasts out of this? BTW this is by far the best styles model, well done good person!
"how on earth do you get SMALL breasts out of this?" ... impossible :D
i'm working on the next version... trying to solve this with loras, but it is a long way
@Guntwin still very exciting model!